AlphaFraud hunts for confident failures, but it is an even-handed analyzer — so we open
with a positive control: a protein AlphaFold had never seen, yet predicted almost
perfectly (the metrics certify it with a near-zero fraud score). Below it, every week we automatically
catch the most striking confidently-wrong structure from that week's PDB release, and beneath the weekly
archive sit our hand-curated deep-dives into the worst offenders.
When AlphaFold gets it right — the control
AlphaFraud exists to catch confident failures — so here is the opposite, on purpose. This is a genuine positive control: a protein AlphaFold had never seen, predicted almost perfectly. It proves two things at once — that AlphaFold can be brilliantly accurate, and that AlphaFraud's metrics certify a good prediction as good (near-zero fraud) rather than only finding bad ones.
Protein Artemis DCLRE1C
Excellent match — AlphaFold nailed itnovelwell-calibrated1.5 Å X-rayQ96SD1 ↗
AlphaFold's blind prediction of a novel, disease-critical nuclease matched the 1.5 Å crystal structure to 0.45 Å RMSD — the positive control that shows the tool is genuinely accurate, and that AlphaFraud certifies good predictions, not just bad ones.
Worst deposition: 6TT5_AAA ·
drag to rotate · scroll to zoom · toggle the AlphaFold “ghost” to see what it predicted.
0.996
TM-score
0.985
lDDT
95.16
mean pLDDT
0.45
Cα-RMSD (Å)
0.02
FRAUD score
71%
novelty
🧬 Structure & architecture
Artemis (gene DCLRE1C, UniProt Q96SD1) is a structure-specific endonuclease whose catalytic core belongs to the metallo-β-lactamase (MBL) fold, fused to a β-CASP domain. This MBL/β-CASP combination defines a nuclease superfamily shared with SNM1A, SNM1B/Apollo and CPSF73, in which the two domains pack together to build a single composite active site rather than acting as independent modules. The MBL domain contributes a shallow αβ/βα sandwich that cradles catalytic metal ions (Zn²⁺), while the β-CASP insertion supplies additional coordinating residues and closes over the metal centre to form a deep, substrate-binding cleft suited to nicking hairpin and branched DNA.
PDB 6TT5 captures this catalytic domain (roughly the first 361 residues) at 1.5 Å resolution — a well-ordered, compact single-domain fold with no oligomeric assembly, no domain-swapping and no conformational ambiguity in the crystal. Full-length Artemis extends well beyond this core with a long C-terminal region that is intrinsically disordered and regulatory, phosphorylated by DNA-PKcs and ATM. That tail is not part of the crystallised, folded domain modelled here — and it is exactly the kind of flexible extension AlphaFold typically flags with low pLDDT rather than mis-predicting with false confidence.
✅ Why AlphaFold gets it right
Every headline metric on this entry points the same way. Global superposition of AlphaFold's blind model against the 6TT5 experimental structure gives a TM-score of 0.996 and GDT_TS of 99.2 (GDT_HA 96.0), with a Cα-RMSD of just 0.45 Å across the full catalytic domain. Locally the agreement is equally tight: lDDT 0.985 and secondary-structure agreement Q3 92.5%, meaning essentially every strand, helix and loop is placed and oriented correctly — not just the domain's rough envelope. This was achieved on a sequence with only 29% identity to the closest structure in AlphaFold's pre-cutoff training set (6TT5 was deposited in 2019, after AlphaFold2's 2018-04-30 cutoff), so the model was not recalling a near-identical template — it was genuinely generalising from evolutionary and physical constraints.
Confidence and correctness are also well matched here, which is the part AlphaFraud is built to check. Mean pLDDT was 95.2, and that confidence was earned: the pLDDT↔lDDT correlation is +0.56, the PAE-overconfident fraction is 0.00, and the confidently-wrong-residue fraction is 0.00 — nowhere did the model report high confidence in a region it actually got wrong. Structurally this is the regime AlphaFold is built for: a single, compact, evolutionarily well-conserved globular enzyme domain with a deep multiple-sequence alignment, one dominant native conformation, and none of the hinge motion, domain-swapping, assembly or amyloid ambiguity that trips it up elsewhere. AlphaFraud's composite score reflects that cleanly: FRAUD = 0.020, against a backdrop where the site's worst offenders score 0.8–0.95. The pipeline is not tuned to find fraud everywhere; here it correctly recognises that there is none.
🩺 Biology & disease
Artemis sits at the centre of non-homologous end joining (NHEJ), the dominant pathway for repairing DNA double-strand breaks in mammalian cells. Recruited to DNA ends by DNA-PKcs and activated by DNA-PKcs-mediated phosphorylation, its endonuclease activity opens the covalently sealed hairpin intermediates generated during V(D)J recombination — the process that assembles diverse antigen-receptor genes in developing T and B lymphocytes. Beyond hairpin opening it also trims and resects DNA ends at double-strand breaks and during immunoglobulin class-switch recombination, making it essential both for adaptive-immune diversity and for genome stability after damage.
Loss-of-function mutations in DCLRE1C cause radiosensitive severe combined immunodeficiency (RS-SCID), including the founder Athabascan SCID phenotype, and Omenn syndrome; hypomorphic alleles are linked to milder combined immunodeficiency, elevated lymphoma risk and clinical radiosensitivity. Because Artemis is required to repair a defined subset of radiation-induced breaks, and because tumour cells often rely on NHEJ to survive genotoxic therapy, it is actively pursued as a radiosensitiser and anti-cancer drug target; the structure reported alongside 6TT5 was explicitly generated to support structure-based strategies for its inhibition.
🎣 Fresh catch — example of the week 2026-07-22
Auto-selected from this release's confidently-wrong catches and written deterministically
from the database — no hand-curation.
✨ NEW
Transcription initiation factor TFIID subunit 8
Right fold, wrong conformation/assemblyassembly🤖 autoQ7Z7C8 ↗
AlphaFold modelled 30FF_DH at mean pLDDT 87.9, but the deposited structure scores TM-score 0.48 against that prediction — 98% of residues confidently wrong. A right fold, wrong conformation/assembly.
Worst deposition: 30FF_DH ·
drag to rotate · scroll to zoom · toggle the AlphaFold “ghost” to see what it predicted.
0.48
worst TM-score
87.93
mean pLDDT
0.82
FRAUD score
×29
structures
27/29
conf. wrong
0%
novelty
🧬 Structure & architecture
Transcription initiation factor TFIID subunit 8 is a 310-residue chain; the deposition resolves 59 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
❌ Why AlphaFold gets it wrong
AlphaFold's mean pLDDT of 87.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 27.62 Å, lDDT 0.85, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.73. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
🗄️ Weekly archive — 457 release weeks
Every weekly PDB release since AlphaFold's cutoff. 347 produced a
confidently-wrong catch (expand to read its panel); the rest were quiet weeks where
AlphaFold was never overconfident — no fraud to report.
AlphaFold modelled 9T5H_M at mean pLDDT 88.5, but the deposited structure scores TM-score 0.42 against that prediction — 96% of residues confidently wrong. A right fold, wrong conformation/assembly.
DNA-directed RNA polymerase II subunit GRINL1A is a 368-residue chain; the deposition resolves 14 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 15.34 Å, lDDT 0.87, secondary-structure agreement Q3 89%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.63. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9RQM_B at mean pLDDT 90.6, but the deposited structure scores TM-score 0.29 against that prediction — 95% of residues confidently wrong. A wrong biological context.
Plexin-B2 is a 25-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 11.53 Å, lDDT 0.53, secondary-structure agreement Q3 59%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.10. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2026-07-01Putative ribosome-binding factor A, mitochondrialWrong biological contextTM 0.20pLDDT 83FRAUD 0.64
AlphaFold modelled 9ROT_a at mean pLDDT 83.1, but the deposited structure scores TM-score 0.20 against that prediction — 77% of residues confidently wrong. A wrong biological context.
Putative ribosome-binding factor A, mitochondrial is a 343-residue chain; the deposition resolves 33 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.20, Cα-RMSD 22.84 Å, lDDT 0.65, secondary-structure agreement Q3 45%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.55. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2026-06-24Keratin, type I cytoskeletal 14Disorder / coiled-coil ambiguityTM 0.33pLDDT 91FRAUD 0.38
AlphaFold modelled 9VHS_A at mean pLDDT 91.4, but the deposited structure scores TM-score 0.33 against that prediction — 76% of residues confidently wrong. A disorder / coiled-coil ambiguity.
Keratin, type I cytoskeletal 14 is a 25-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.4 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.33, Cα-RMSD 6.72 Å, lDDT 0.52, secondary-structure agreement Q3 28%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.08. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 11MG_B at mean pLDDT 79.4, but the deposited structure scores TM-score 0.45 against that prediction — 78% of residues confidently wrong. A wrong biological context.
Complement C3b alpha' chain is a 915-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 29.18 Å, lDDT 0.76, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.59. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9YQP_A at mean pLDDT 85.5, but the deposited structure scores TM-score 0.45 against that prediction — 87% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 149-residue chain annotated as EF-hand; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 12.23 Å, lDDT 0.80, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.43. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9RX6_D at mean pLDDT 86.1, but the deposited structure scores TM-score 0.39 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
UV radiation resistance associated protein is a 699-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.39, Cα-RMSD 15.46 Å, lDDT 0.79, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.38. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9OI9_A at mean pLDDT 88.7, but the deposited structure scores TM-score 0.23 against that prediction — 81% of residues confidently wrong. A wrong biological context.
ADan (ITM2B mutant) is a 25-residue chain; the deposition resolves 15 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.23, Cα-RMSD 6.60 Å, lDDT 0.52, secondary-structure agreement Q3 43%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.25. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2026-05-20GATOR2 complex protein WDR24Right fold, wrong conformation/assemblyTM 0.43pLDDT 86FRAUD 0.84
AlphaFold modelled 9DX1_F at mean pLDDT 86.1, but the deposited structure scores TM-score 0.43 against that prediction — 85% of residues confidently wrong. A right fold, wrong conformation/assembly.
GATOR2 complex protein WDR24 is a 729-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 41.36 Å, lDDT 0.77, secondary-structure agreement Q3 72%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.70. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 74% novel to AlphaFold's training cutoff.
2026-05-13B-cell CLL/lymphoma 7 protein family member AWrong biological contextTM 0.44pLDDT 88FRAUD 0.37
AlphaFold modelled 21VV_N at mean pLDDT 87.9, but the deposited structure scores TM-score 0.44 against that prediction — 59% of residues confidently wrong. A wrong biological context.
B-cell CLL/lymphoma 7 protein family member A is a 216-residue chain; the deposition resolves 14 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 7.34 Å, lDDT 0.80, secondary-structure agreement Q3 50%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.20. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2026-05-06B-cell CLL/lymphoma 7 protein family member BWrong biological contextTM 0.40pLDDT 88FRAUD 0.36
AlphaFold modelled 9UXA_L at mean pLDDT 87.7, but the deposited structure scores TM-score 0.40 against that prediction — 64% of residues confidently wrong. A wrong biological context.
B-cell CLL/lymphoma 7 protein family member B is a 203-residue chain; the deposition resolves 14 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 6.96 Å, lDDT 0.81, secondary-structure agreement Q3 57%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.14. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2026-04-29Structural maintenance of chromosomes protein 6Right fold, wrong conformation/assemblyTM 0.43pLDDT 83FRAUD 0.83
AlphaFold modelled 9LWL_A at mean pLDDT 82.9, but the deposited structure scores TM-score 0.43 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
Structural maintenance of chromosomes protein 6 is a 1091-residue chain; the deposition resolves 10 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 122.25 Å, lDDT 0.78, secondary-structure agreement Q3 89%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.46. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 71% novel to AlphaFold's training cutoff.
AlphaFold modelled 11TA_I at mean pLDDT 95.1, but the deposited structure scores TM-score 0.39 against that prediction — 100% of residues confidently wrong. A right fold, wrong conformation/assembly.
Ubiquitin is a 76-residue chain; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 95.1 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.39, Cα-RMSD 16.95 Å, lDDT 0.74, secondary-structure agreement Q3 60%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.26. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9UG2_A at mean pLDDT 91.5, but the deposited structure scores TM-score 0.34 against that prediction — 95% of residues confidently wrong. A native fold → amyloid.
Gelsolin is a 348-residue chain; the deposition resolves 9 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.5 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.34, Cα-RMSD 23.30 Å, lDDT 0.80, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.56. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
2026-04-08MyoD family inhibitor domain-containing protein 2Wrong biological contextTM 0.49pLDDT 72FRAUD 0.07
AlphaFold modelled 9VEE_D at mean pLDDT 72.4, but the deposited structure scores TM-score 0.49 against that prediction — 0% of residues confidently wrong. A wrong biological context.
MyoD family inhibitor domain-containing protein 2 is a 211-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.4 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 2.21 Å, lDDT 0.86, secondary-structure agreement Q3 76%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.75. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2026-04-01Transmembrane gamma-carboxyglutamic acid protein 1Wrong biological contextTM 0.26pLDDT 78FRAUD 0.25
AlphaFold modelled 9WFL_B at mean pLDDT 78.2, but the deposited structure scores TM-score 0.26 against that prediction — 48% of residues confidently wrong. A wrong biological context.
Transmembrane gamma-carboxyglutamic acid protein 1 is a 29-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.26, Cα-RMSD 5.35 Å, lDDT 0.71, secondary-structure agreement Q3 76%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.31. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9NUB_A at mean pLDDT 80.9, but the deposited structure scores TM-score 0.39 against that prediction — 73% of residues confidently wrong. A right fold, wrong conformation/assembly.
Matrix Gla protein is a 84-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 80.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.39, Cα-RMSD 13.38 Å, lDDT 0.68, secondary-structure agreement Q3 80%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.49. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9P2Q_A at mean pLDDT 89.3, but the deposited structure scores TM-score 0.44 against that prediction — 23% of residues confidently wrong. A right fold, wrong conformation/assembly.
E3 ubiquitin-protein ligase makorin-3 is a 31-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 3.28 Å, lDDT 0.70, secondary-structure agreement Q3 71%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.38. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9M76_K at mean pLDDT 87.7, but the deposited structure scores TM-score 0.42 against that prediction — 89% of residues confidently wrong. A right fold, wrong conformation/assembly.
E3 ubiquitin-protein ligase UHRF1 is a 793-residue chain annotated as PUA domain-like; the deposition resolves 11 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 21.08 Å, lDDT 0.83, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.60. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2026-03-04DNA ligase 4✓ no confident missTM 0.54pLDDT 91FRAUD 0.89
479 structures analysed this release. AlphaFold's largest disagreement was DNA ligase 4 (TM 0.54) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 9Z44_D at mean pLDDT 88.7, but the deposited structure scores TM-score 0.41 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
SLC2A4 regulator is a 33-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 0.78 Å, lDDT 0.80, secondary-structure agreement Q3 100%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.43. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2026-02-18B-cell CLL/lymphoma 7 protein family member ARight fold, wrong conformation/assemblyTM 0.46pLDDT 90FRAUD 0.25
AlphaFold modelled 9WBZ_O at mean pLDDT 90.3, but the deposited structure scores TM-score 0.46 against that prediction — 39% of residues confidently wrong. A right fold, wrong conformation/assembly.
B-cell CLL/lymphoma 7 protein family member A is a 210-residue chain; the deposition resolves 14 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.3 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 4.88 Å, lDDT 0.82, secondary-structure agreement Q3 61%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.46. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9NON_A at mean pLDDT 97.3, but the deposited structure scores TM-score 0.17 against that prediction — 100% of residues confidently wrong. A native fold → amyloid.
Leukocyte cell-derived chemotaxin-2 is a 133-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 97.3 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.17, Cα-RMSD 21.93 Å, lDDT 0.46, secondary-structure agreement Q3 59%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.09. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
2026-02-04Alpha-synuclein✓ no confident missTM 0.24pLDDT 83FRAUD 0.82
309 structures analysed this release. AlphaFold's largest disagreement was Alpha-synuclein (TM 0.24) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2026-01-28DET1- and DDB1-associated protein 1Wrong biological contextTM 0.31pLDDT 78FRAUD 0.62
AlphaFold modelled 9LUL_L at mean pLDDT 77.6, but the deposited structure scores TM-score 0.31 against that prediction — 94% of residues confidently wrong. A wrong biological context.
DET1- and DDB1-associated protein 1 is a 66-residue chain annotated as UBC-like; the deposition resolves 16 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.31, Cα-RMSD 14.62 Å, lDDT 0.77, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.74. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2026-01-21COMM domain-containing protein 4Right fold, wrong conformation/assemblyTM 0.49pLDDT 83FRAUD 0.51
AlphaFold modelled 9ZMZ_A at mean pLDDT 82.9, but the deposited structure scores TM-score 0.49 against that prediction — 93% of residues confidently wrong. A right fold, wrong conformation/assembly.
COMM domain-containing protein 4 is a 112-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 13.06 Å, lDDT 0.78, secondary-structure agreement Q3 94%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.50. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9ZRK_E at mean pLDDT 86.0, but the deposited structure scores TM-score 0.48 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 146-residue chain annotated as EF-hand; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 9.80 Å, lDDT 0.81, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.48. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2026-01-07Guanine nucleotide-binding protein G(i) subunit alpha-1✓ no confident missTM 0.56pLDDT 94FRAUD 0.73
56 structures analysed this release. AlphaFold's largest disagreement was Guanine nucleotide-binding protein G(i) subunit alpha-1 (TM 0.56) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 9SZT_e at mean pLDDT 73.7, but the deposited structure scores TM-score 0.37 against that prediction — 46% of residues confidently wrong. A wrong biological context.
26S proteasome complex subunit SEM1 is a 70-residue chain; the deposition resolves 47 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 73.7 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 9.28 Å, lDDT 0.69, secondary-structure agreement Q3 62%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.38. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9ZPC_C at mean pLDDT 90.0, but the deposited structure scores TM-score 0.45 against that prediction — 96% of residues confidently wrong. A right fold, wrong conformation/assembly.
UV radiation resistance-associated gene protein is a 699-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.0 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 17.43 Å, lDDT 0.89, secondary-structure agreement Q3 89%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.47. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2025-12-17Ubiquitin-60S ribosomal protein L40Wrong biological contextTM 0.33pLDDT 92FRAUD 0.48
AlphaFold modelled 9SYR_LD at mean pLDDT 92.0, but the deposited structure scores TM-score 0.33 against that prediction — 91% of residues confidently wrong. A wrong biological context.
Ubiquitin-60S ribosomal protein L40 is a 99-residue chain; the deposition resolves 86 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 92.0 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.33, Cα-RMSD 8.26 Å, lDDT 0.34, secondary-structure agreement Q3 28%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.12. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9HXW_A at mean pLDDT 79.7, but the deposited structure scores TM-score 0.34 against that prediction — 84% of residues confidently wrong. A wrong biological context.
E3 ubiquitin-protein ligase UBR4 is a 4781-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.34, Cα-RMSD 26.71 Å, lDDT 0.84, secondary-structure agreement Q3 90%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.76. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9LGG_A at mean pLDDT 71.6, but the deposited structure scores TM-score 0.47 against that prediction — 57% of residues confidently wrong. A wrong biological context.
Muscle, skeletal receptor tyrosine-protein kinase,Ig3-Fz domains of hMuSK:L1 peptide is a 279-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 71.6 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 14.53 Å, lDDT 0.70, secondary-structure agreement Q3 87%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.88. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8ZJZ_A at mean pLDDT 87.3, but the deposited structure scores TM-score 0.43 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
Butyrophilin subfamily 2 member A1 is a 492-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 65.67 Å, lDDT 0.85, secondary-structure agreement Q3 91%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.63. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9N9J_G at mean pLDDT 80.6, but the deposited structure scores TM-score 0.45 against that prediction — 79% of residues confidently wrong. A right fold, wrong conformation/assembly.
Stress-associated endoplasmic reticulum protein 1 is a 66-residue chain annotated as SecY-like; the deposition resolves 70 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 80.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 7.70 Å, lDDT 0.81, secondary-structure agreement Q3 86%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.79. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2025-11-12KICSTOR complex protein SZT2Native fold → amyloidTM 0.24pLDDT 85FRAUD 0.80
AlphaFold modelled 9V86_A at mean pLDDT 85.4, but the deposited structure scores TM-score 0.24 against that prediction — 92% of residues confidently wrong. A native fold → amyloid.
KICSTOR complex protein SZT2 is a 3432-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.24, Cα-RMSD 21.94 Å, lDDT 0.20, secondary-structure agreement Q3 42%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.21. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9UC8_B at mean pLDDT 86.3, but the deposited structure scores TM-score 0.40 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 149-residue chain annotated as EF-hand; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 12.90 Å, lDDT 0.60, secondary-structure agreement Q3 79%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.08. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9K8S_E at mean pLDDT 76.8, but the deposited structure scores TM-score 0.17 against that prediction — 76% of residues confidently wrong. A native fold → amyloid.
Ninjurin-1 is a 30-residue chain; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.17, Cα-RMSD 14.05 Å, lDDT 0.41, secondary-structure agreement Q3 0%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.26. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9L24_B at mean pLDDT 79.7, but the deposited structure scores TM-score 0.24 against that prediction — 79% of residues confidently wrong. A wrong biological context.
Matrix Gla protein is a 29-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.24, Cα-RMSD 8.49 Å, lDDT 0.65, secondary-structure agreement Q3 59%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.60. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9JXV_E at mean pLDDT 79.6, but the deposited structure scores TM-score 0.50 against that prediction — 47% of residues confidently wrong. A wrong biological context.
Mitochondrial import receptor subunit TOM6 homolog is a 74-residue chain; the deposition resolves 10 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 6.43 Å, lDDT 0.81, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.68. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9MQE_P at mean pLDDT 81.8, but the deposited structure scores TM-score 0.26 against that prediction — 60% of residues confidently wrong. A wrong biological context.
Osteocalcin is a 100-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.26, Cα-RMSD 16.96 Å, lDDT 0.57, secondary-structure agreement Q3 49%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.06. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9DT0_A at mean pLDDT 81.4, but the deposited structure scores TM-score 0.38 against that prediction — 76% of residues confidently wrong. A disorder / coiled-coil ambiguity.
Small EDRK-rich factor 2 is a 59-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 13.45 Å, lDDT 0.63, secondary-structure agreement Q3 47%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.15. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9S0U_V at mean pLDDT 81.2, but the deposited structure scores TM-score 0.39 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
RNA polymerase II-associated factor 1 homolog is a 531-residue chain annotated as Core histone-like; the deposition resolves 35 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.39, Cα-RMSD 40.46 Å, lDDT 0.63, secondary-structure agreement Q3 67%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.14. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9Y0R_B at mean pLDDT 96.7, but the deposited structure scores TM-score 0.33 against that prediction — 75% of residues confidently wrong. A wrong biological context.
Cyclic AMP-dependent transcription factor ATF-2 is a 27-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.7 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.33, Cα-RMSD 8.81 Å, lDDT 0.57, secondary-structure agreement Q3 0%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.58. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2025-09-15Metalloproteinase inhibitor 1✓ no confident missTM 0.94pLDDT 90FRAUD 0.06
1 structure analysed this release. AlphaFold's largest disagreement was Metalloproteinase inhibitor 1 (TM 0.94) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 9M2W_e at mean pLDDT 70.9, but the deposited structure scores TM-score 0.29 against that prediction — 42% of residues confidently wrong. A wrong biological context.
26S proteasome complex subunit SEM1 is a 70-residue chain; the deposition resolves 48 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 70.9 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 8.84 Å, lDDT 0.57, secondary-structure agreement Q3 24%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.43. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9QWZ_B at mean pLDDT 76.5, but the deposited structure scores TM-score 0.35 against that prediction — 72% of residues confidently wrong. A wrong biological context.
E3 ubiquitin-protein ligase UBR4 is a 5193-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 19.86 Å, lDDT 0.64, secondary-structure agreement Q3 66%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.72. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8VXJ_C at mean pLDDT 75.8, but the deposited structure scores TM-score 0.30 against that prediction — 64% of residues confidently wrong. A wrong biological context.
Apolipoprotein A-I is a 243-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 75.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.30, Cα-RMSD 35.46 Å, lDDT 0.38, secondary-structure agreement Q3 78%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.55. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9GFB_H at mean pLDDT 81.9, but the deposited structure scores TM-score 0.38 against that prediction — 81% of residues confidently wrong. A right fold, wrong conformation/assembly.
INO80 complex subunit B is a 356-residue chain; the deposition resolves 20 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 32.42 Å, lDDT 0.89, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.59. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9D1V_A at mean pLDDT 93.6, but the deposited structure scores TM-score 0.47 against that prediction — 93% of residues confidently wrong. A right fold, wrong conformation/assembly.
Beta-crystallin B2 is a 205-residue chain annotated as gamma-Crystallin-like; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 93.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 18.48 Å, lDDT 0.91, secondary-structure agreement Q3 89%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.67. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9P3Z_A at mean pLDDT 87.0, but the deposited structure scores TM-score 0.37 against that prediction — 37% of residues confidently wrong. A right fold, wrong conformation/assembly.
Zinc finger C4H2 domain-containing protein is a 27-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 3.90 Å, lDDT 0.61, secondary-structure agreement Q3 74%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.51. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9RZF_A at mean pLDDT 81.0, but the deposited structure scores TM-score 0.29 against that prediction — 84% of residues confidently wrong. A native fold → amyloid.
Alpha-synuclein is a 61-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 19.45 Å, lDDT 0.31, secondary-structure agreement Q3 20%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.46. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9O55_C at mean pLDDT 96.5, but the deposited structure scores TM-score 0.33 against that prediction — 50% of residues confidently wrong. A wrong biological context.
GTPase KRas, N-terminally processed is a 10-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.5 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.33, Cα-RMSD 4.42 Å, lDDT 0.67, secondary-structure agreement Q3 50%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.26. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9G0A_Z at mean pLDDT 90.0, but the deposited structure scores TM-score 0.41 against that prediction — 98% of residues confidently wrong. A right fold, wrong conformation/assembly.
Transcription elongation factor SPT5 is a 1087-residue chain; the deposition resolves 23 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.0 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 21.08 Å, lDDT 0.85, secondary-structure agreement Q3 83%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.14. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2025-07-09Vacuolar protein sorting-associated protein 72 homologRight fold, wrong conformation/assemblyTM 0.38pLDDT 85FRAUD 0.36
AlphaFold modelled 9INC_C at mean pLDDT 85.3, but the deposited structure scores TM-score 0.38 against that prediction — 72% of residues confidently wrong. A right fold, wrong conformation/assembly.
Vacuolar protein sorting-associated protein 72 homolog is a 87-residue chain annotated as YL1-Z domain-like; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 7.15 Å, lDDT 0.86, secondary-structure agreement Q3 70%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.03. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9F5W_H at mean pLDDT 78.4, but the deposited structure scores TM-score 0.36 against that prediction — 73% of residues confidently wrong. A wrong biological context.
Condensin-2 complex subunit H2 is a 640-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 39.76 Å, lDDT 0.71, secondary-structure agreement Q3 87%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.41. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9VD9_A at mean pLDDT 81.5, but the deposited structure scores TM-score 0.45 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
Integrator complex subunit 1 is a 2190-residue chain; the deposition resolves 38 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 56.43 Å, lDDT 0.76, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.53. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9JQQ_E at mean pLDDT 88.5, but the deposited structure scores TM-score 0.42 against that prediction — 94% of residues confidently wrong. A right fold, wrong conformation/assembly.
Butyrophilin subfamily 2 member A1 is a 532-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 67.95 Å, lDDT 0.84, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.44. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9D9Z_A at mean pLDDT 80.5, but the deposited structure scores TM-score 0.35 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
UBR4 (endogenously FLAG-tagged at the N-terminus),E3 ubiquitin-protein ligase UBR4,E3 ubiquitin-protein ligase UBR4,E3 ubiquitin-protein ligase UBR4 is a 5205-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 80.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 26.31 Å, lDDT 0.83, secondary-structure agreement Q3 91%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.71. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8QVR_A at mean pLDDT 76.2, but the deposited structure scores TM-score 0.28 against that prediction — 64% of residues confidently wrong. A native fold → amyloid.
Islet amyloid polypeptide is a 38-residue chain; the deposition resolves 10 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.28, Cα-RMSD 11.67 Å, lDDT 0.37, secondary-structure agreement Q3 42%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.36. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
2025-05-285'-nucleotidase✓ no confident missTM 0.69pLDDT 96FRAUD 0.49
111 structures analysed this release. AlphaFold's largest disagreement was 5'-nucleotidase (TM 0.69) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 9J5J_E at mean pLDDT 87.3, but the deposited structure scores TM-score 0.43 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
Butyrophilin subfamily 2 member A1 is a 499-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 70.77 Å, lDDT 0.82, secondary-structure agreement Q3 89%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.59. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9NZH_B at mean pLDDT 76.4, but the deposited structure scores TM-score 0.46 against that prediction — 62% of residues confidently wrong. A native fold → amyloid.
Islet amyloid polypeptide is a 37-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 8.89 Å, lDDT 0.60, secondary-structure agreement Q3 76%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.68. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9O6T_B at mean pLDDT 86.0, but the deposited structure scores TM-score 0.44 against that prediction — 66% of residues confidently wrong. A right fold, wrong conformation/assembly.
Prohibitin 1 is a 272-residue chain; the deposition resolves 24 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 8.04 Å, lDDT 0.80, secondary-structure agreement Q3 78%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.68. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9QIW_CE at mean pLDDT 90.6, but the deposited structure scores TM-score 0.48 against that prediction — 99% of residues confidently wrong. A right fold, wrong conformation/assembly.
Protein LLP homolog is a 129-residue chain; the deposition resolves 51 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 14.62 Å, lDDT 0.90, secondary-structure agreement Q3 94%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.06. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9CQ3_F at mean pLDDT 91.0, but the deposited structure scores TM-score 0.35 against that prediction — 96% of residues confidently wrong. A right fold, wrong conformation/assembly.
DNA ligase 4 is a 914-residue chain annotated as DNA ligase IV Xrcc4-binding domain-like; the deposition resolves 20 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.0 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 42.58 Å, lDDT 0.89, secondary-structure agreement Q3 88%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.28. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9B7J_R at mean pLDDT 89.9, but the deposited structure scores TM-score 0.42 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
Dynactin subunit 3 is a 186-residue chain; the deposition resolves 25 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 12.22 Å, lDDT 0.86, secondary-structure agreement Q3 90%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.66. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2025-04-09Isoform Short of Insulin receptorRight fold, wrong conformation/assemblyTM 0.50pLDDT 88FRAUD 0.75
AlphaFold modelled 8YYT_A at mean pLDDT 87.9, but the deposited structure scores TM-score 0.50 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
Isoform Short of Insulin receptor is a 1370-residue chain annotated as L domain-like; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 18.09 Å, lDDT 0.83, secondary-structure agreement Q3 87%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.41. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9MXZ_A at mean pLDDT 75.8, but the deposited structure scores TM-score 0.44 against that prediction — 70% of residues confidently wrong. A wrong biological context.
Apolipoprotein A-I is a 243-residue chain annotated as Apolipoprotein; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 75.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 21.00 Å, lDDT 0.85, secondary-structure agreement Q3 91%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.78. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8YAD_B at mean pLDDT 72.7, but the deposited structure scores TM-score 0.37 against that prediction — 69% of residues confidently wrong. A wrong biological context.
Spatacsin is a 2443-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.7 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 93.41 Å, lDDT 0.69, secondary-structure agreement Q3 80%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.40. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8W8Y_L at mean pLDDT 74.6, but the deposited structure scores TM-score 0.25 against that prediction — 27% of residues confidently wrong. A wrong biological context.
Gama-MSH is a 11-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 74.6 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.25, Cα-RMSD 4.23 Å, lDDT 0.62, secondary-structure agreement Q3 73%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.38. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9CKX_A at mean pLDDT 77.5, but the deposited structure scores TM-score 0.16 against that prediction — 60% of residues confidently wrong. A wrong biological context.
Ubiquitin-domain-containing protein,Response regulator FrzS,Vesicle-associated membrane protein 2 is a 222-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.16, Cα-RMSD 41.52 Å, lDDT 0.41, secondary-structure agreement Q3 62%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.72. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2025-03-05DET1- and DDB1-associated protein 1Wrong biological contextTM 0.36pLDDT 77FRAUD 0.57
AlphaFold modelled 9BJZ_C at mean pLDDT 76.8, but the deposited structure scores TM-score 0.36 against that prediction — 84% of residues confidently wrong. A wrong biological context.
DET1- and DDB1-associated protein 1 is a 102-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 14.02 Å, lDDT 0.74, secondary-structure agreement Q3 88%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.60. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8YTC_A at mean pLDDT 85.6, but the deposited structure scores TM-score 0.30 against that prediction — 89% of residues confidently wrong. A wrong biological context.
Protein PML is a 211-residue chain annotated as RING/U-box like; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.30, Cα-RMSD 13.55 Å, lDDT 0.33, secondary-structure agreement Q3 77%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.04. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9KBD_R at mean pLDDT 85.3, but the deposited structure scores TM-score 0.27 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
E3 ubiquitin-protein ligase RBX1 is a 108-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.27, Cα-RMSD 1.68 Å, lDDT 0.90, secondary-structure agreement Q3 81%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.21. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9FNO_C at mean pLDDT 91.9, but the deposited structure scores TM-score 0.48 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
Isoform 2 of Haptoglobin alpha chain is a 83-residue chain; the deposition resolves 7 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.9 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 0.93 Å, lDDT 0.93, secondary-structure agreement Q3 90%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.14. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2025-02-05DET1- and DDB1-associated protein 1Wrong biological contextTM 0.28pLDDT 77FRAUD 0.56
AlphaFold modelled 9DHD_C at mean pLDDT 76.7, but the deposited structure scores TM-score 0.28 against that prediction — 89% of residues confidently wrong. A wrong biological context.
DET1- and DDB1-associated protein 1 is a 102-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.28, Cα-RMSD 12.98 Å, lDDT 0.77, secondary-structure agreement Q3 89%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.56. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9K26_F at mean pLDDT 71.8, but the deposited structure scores TM-score 0.16 against that prediction — 6% of residues confidently wrong. A wrong biological context.
Prolactin-releasing peptide PrRP31 is a 32-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 71.8 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.16, Cα-RMSD 4.36 Å, lDDT 0.70, secondary-structure agreement Q3 100%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.19. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9BVR_P at mean pLDDT 78.2, but the deposited structure scores TM-score 0.32 against that prediction — 62% of residues confidently wrong. A wrong biological context.
Coagulation factor IX is a 64-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.32, Cα-RMSD 9.89 Å, lDDT 0.49, secondary-structure agreement Q3 51%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.11. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8RL2_CE at mean pLDDT 90.6, but the deposited structure scores TM-score 0.48 against that prediction — 99% of residues confidently wrong. A right fold, wrong conformation/assembly.
Protein LLP homolog is a 129-residue chain; the deposition resolves 52 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 14.50 Å, lDDT 0.90, secondary-structure agreement Q3 94%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.01. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
62 structures analysed this release. AlphaFold's largest disagreement was Adenosine receptor A2a,Soluble cytochrome b562 (TM 0.78) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
69 structures analysed this release. AlphaFold's largest disagreement was Speckle targeted PIP5K1A-regulated poly(A) polymerase (TM 0.57) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2024-12-25DET1- and DDB1-associated protein 1Wrong biological contextTM 0.34pLDDT 72FRAUD 0.45
AlphaFold modelled 9BZ0_e at mean pLDDT 71.5, but the deposited structure scores TM-score 0.34 against that prediction — 64% of residues confidently wrong. A wrong biological context.
DET1- and DDB1-associated protein 1 is a 105-residue chain; the deposition resolves 22 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 71.5 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.34, Cα-RMSD 10.52 Å, lDDT 0.68, secondary-structure agreement Q3 70%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.57. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9EGX_V at mean pLDDT 81.1, but the deposited structure scores TM-score 0.38 against that prediction — 85% of residues confidently wrong. A right fold, wrong conformation/assembly.
RNA polymerase II-associated factor 1 homolog is a 531-residue chain; the deposition resolves 32 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 41.14 Å, lDDT 0.57, secondary-structure agreement Q3 69%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.19. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2024-12-11Isoform 4 of Adhesion G protein-coupled receptor L3Wrong biological contextTM 0.45pLDDT 78FRAUD 0.03
AlphaFold modelled 8VTI_B at mean pLDDT 78.3, but the deposited structure scores TM-score 0.45 against that prediction — 0% of residues confidently wrong. A wrong biological context.
Isoform 4 of Adhesion G protein-coupled receptor L3 is a 314-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 0.82 Å, lDDT 0.96, secondary-structure agreement Q3 91%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.63. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 70% novel to AlphaFold's training cutoff.
AlphaFold modelled 9CE3_C at mean pLDDT 84.7, but the deposited structure scores TM-score 0.44 against that prediction — 85% of residues confidently wrong. A right fold, wrong conformation/assembly.
Hamartin is a 1191-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 84.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 75.25 Å, lDDT 0.83, secondary-structure agreement Q3 94%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.60. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2024-11-27Alpha-synuclein✓ no confident missTM 0.24pLDDT 88FRAUD 0.74
145 structures analysed this release. AlphaFold's largest disagreement was Alpha-synuclein (TM 0.24) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8R4A_A at mean pLDDT 98.4, but the deposited structure scores TM-score 0.23 against that prediction — 98% of residues confidently wrong. A native fold → amyloid.
Lysozyme C is a 130-residue chain annotated as Lysozyme-like; the deposition resolves 9 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 98.4 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.23, Cα-RMSD 21.98 Å, lDDT 0.49, secondary-structure agreement Q3 69%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.37. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9CUI_E at mean pLDDT 85.5, but the deposited structure scores TM-score 0.50 against that prediction — 79% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 149-residue chain annotated as EF-hand; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 12.82 Å, lDDT 0.79, secondary-structure agreement Q3 91%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.58. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8USC_e at mean pLDDT 75.6, but the deposited structure scores TM-score 0.37 against that prediction — 56% of residues confidently wrong. A wrong biological context.
26S proteasome complex subunit SEM1 is a 70-residue chain; the deposition resolves 28 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 75.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 8.34 Å, lDDT 0.67, secondary-structure agreement Q3 61%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.61. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8IHU_A at mean pLDDT 98.3, but the deposited structure scores TM-score 0.26 against that prediction — 97% of residues confidently wrong. A native fold → amyloid.
Superoxide dismutase [Cu-Zn] is a 154-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 98.3 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.26, Cα-RMSD 17.91 Å, lDDT 0.45, secondary-structure agreement Q3 61%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.34. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9G8M_Sf at mean pLDDT 90.4, but the deposited structure scores TM-score 0.42 against that prediction — 82% of residues confidently wrong. A wrong biological context.
Ubiquitin is a 156-residue chain; the deposition resolves 92 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.4 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 6.79 Å, lDDT 0.54, secondary-structure agreement Q3 82%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.24. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8VIZ_G at mean pLDDT 92.9, but the deposited structure scores TM-score 0.34 against that prediction — 97% of residues confidently wrong. A native fold → amyloid.
Gelsolin is a 746-residue chain annotated as Actin depolymerizing proteins; the deposition resolves 7 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 92.9 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.34, Cα-RMSD 33.10 Å, lDDT 0.76, secondary-structure agreement Q3 88%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.40. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9JST_A at mean pLDDT 91.2, but the deposited structure scores TM-score 0.29 against that prediction — 97% of residues confidently wrong. A native fold → amyloid.
M-alpha is a 35-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.2 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 9.00 Å, lDDT 0.62, secondary-structure agreement Q3 57%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.18. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9C3E_X at mean pLDDT 81.2, but the deposited structure scores TM-score 0.46 against that prediction — 43% of residues confidently wrong. A right fold, wrong conformation/assembly.
T-cell surface glycoprotein CD3 zeta chain is a 164-residue chain; the deposition resolves 9 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 5.09 Å, lDDT 0.79, secondary-structure agreement Q3 75%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.52. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8ZPT_L at mean pLDDT 70.9, but the deposited structure scores TM-score 0.23 against that prediction — 25% of residues confidently wrong. A wrong biological context.
Prolactin-releasing peptide PrRP20 is a 21-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 70.9 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.23, Cα-RMSD 5.20 Å, lDDT 0.70, secondary-structure agreement Q3 100%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.07. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2024-09-18Coiled-coil domain-containing protein 12Right fold, wrong conformation/assemblyTM 0.41pLDDT 83FRAUD 0.81
AlphaFold modelled 8RO2_Z at mean pLDDT 82.9, but the deposited structure scores TM-score 0.41 against that prediction — 84% of residues confidently wrong. A right fold, wrong conformation/assembly.
Coiled-coil domain-containing protein 12 is a 166-residue chain; the deposition resolves 40 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 22.55 Å, lDDT 0.86, secondary-structure agreement Q3 92%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.65. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8W2V_A at mean pLDDT 79.0, but the deposited structure scores TM-score 0.32 against that prediction — 56% of residues confidently wrong. A disorder / coiled-coil ambiguity.
T-cell-specific surface glycoprotein CD28 is a 39-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.32, Cα-RMSD 7.45 Å, lDDT 0.54, secondary-structure agreement Q3 59%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.15. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9A3F_C at mean pLDDT 91.3, but the deposited structure scores TM-score 0.48 against that prediction — 86% of residues confidently wrong. A wrong biological context.
GA is a 394-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.3 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 19.36 Å, lDDT 0.49, secondary-structure agreement Q3 75%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.40. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8VRA_C at mean pLDDT 96.5, but the deposited structure scores TM-score 0.28 against that prediction — 20% of residues confidently wrong. A wrong biological context.
GTPase KRas, N-terminally processed is a 10-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.5 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.28, Cα-RMSD 4.34 Å, lDDT 0.65, secondary-structure agreement Q3 50%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.09. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8XVH_T at mean pLDDT 77.1, but the deposited structure scores TM-score 0.38 against that prediction — 0% of residues confidently wrong. A wrong biological context.
Endothelin-1 is a 21-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 3.55 Å, lDDT 0.81, secondary-structure agreement Q3 81%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.83. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2024-08-21Vacuolar protein sorting-associated protein 72 homologRight fold, wrong conformation/assemblyTM 0.32pLDDT 90FRAUD 0.89
AlphaFold modelled 9C62_K at mean pLDDT 89.5, but the deposited structure scores TM-score 0.32 against that prediction — 98% of residues confidently wrong. A right fold, wrong conformation/assembly.
Vacuolar protein sorting-associated protein 72 homolog is a 364-residue chain; the deposition resolves 15 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.32, Cα-RMSD 30.97 Å, lDDT 0.84, secondary-structure agreement Q3 82%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.24. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9GAW_M at mean pLDDT 76.1, but the deposited structure scores TM-score 0.25 against that prediction — 61% of residues confidently wrong. A wrong biological context.
Anaphase-promoting complex subunit 13 is a 74-residue chain; the deposition resolves 21 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.25, Cα-RMSD 21.30 Å, lDDT 0.72, secondary-structure agreement Q3 80%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.67. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8XVA_K at mean pLDDT 76.4, but the deposited structure scores TM-score 0.36 against that prediction — 72% of residues confidently wrong. A wrong biological context.
Mitochondrial import receptor subunit TOM20 homolog is a 145-residue chain; the deposition resolves 11 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 17.67 Å, lDDT 0.71, secondary-structure agreement Q3 83%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.11. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8I0S_v at mean pLDDT 89.9, but the deposited structure scores TM-score 0.40 against that prediction — 100% of residues confidently wrong. A right fold, wrong conformation/assembly.
Splicing factor 3A subunit 2 is a 464-residue chain; the deposition resolves 50 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 23.00 Å, lDDT 0.74, secondary-structure agreement Q3 71%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.01. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 72% novel to AlphaFold's training cutoff.
AlphaFold modelled 9BQ2_B at mean pLDDT 82.2, but the deposited structure scores TM-score 0.41 against that prediction — 78% of residues confidently wrong. A right fold, wrong conformation/assembly.
Flotillin-1 is a 422-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 20.27 Å, lDDT 0.86, secondary-structure agreement Q3 91%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.62. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2024-07-17Zinc finger protein 711Whole-fold mismatchTM 0.46pLDDT 74FRAUD 0.09
AlphaFold modelled 9CJA_A at mean pLDDT 73.7, but the deposited structure scores TM-score 0.46 against that prediction — 0% of residues confidently wrong. A whole-fold mismatch.
Zinc finger protein 711 is a 29-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 73.7 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 2.52 Å, lDDT 0.76, secondary-structure agreement Q3 100%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.74. Mechanistically, secondary structure and tertiary packing both diverge from the prediction — AlphaFold committed to the wrong fold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9FMD_x at mean pLDDT 78.0, but the deposited structure scores TM-score 0.27 against that prediction — 77% of residues confidently wrong. A wrong biological context.
Splicing factor C9orf78 is a 289-residue chain; the deposition resolves 63 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.27, Cα-RMSD 41.43 Å, lDDT 0.90, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.57. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2024-07-03Mediator of RNA polymerase II transcription subunit 11Right fold, wrong conformation/assemblyTM 0.44pLDDT 89FRAUD 0.51
AlphaFold modelled 8T9D_H at mean pLDDT 88.8, but the deposited structure scores TM-score 0.44 against that prediction — 78% of residues confidently wrong. A right fold, wrong conformation/assembly.
Mediator of RNA polymerase II transcription subunit 11 is a 117-residue chain; the deposition resolves 26 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 10.03 Å, lDDT 0.74, secondary-structure agreement Q3 61%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.53. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2024-06-26Engulfment and cell motility protein 1✓ no confident missTM 0.69pLDDT 89FRAUD 0.74
351 structures analysed this release. AlphaFold's largest disagreement was Engulfment and cell motility protein 1 (TM 0.69) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8WZ2_L at mean pLDDT 71.0, but the deposited structure scores TM-score 0.43 against that prediction — 40% of residues confidently wrong. A wrong biological context.
Orexigenic neuropeptide QRFP is a 27-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 71.0 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 6.24 Å, lDDT 0.78, secondary-structure agreement Q3 72%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.41. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2024-06-12Plasminogen✓ no confident missTM 0.58pLDDT 85FRAUD 0.68
246 structures analysed this release. AlphaFold's largest disagreement was Plasminogen (TM 0.58) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2024-06-05Alpha-synuclein✓ no confident missTM 0.24pLDDT 85FRAUD 0.81
90 structures analysed this release. AlphaFold's largest disagreement was Alpha-synuclein (TM 0.24) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 9BV0_A at mean pLDDT 82.6, but the deposited structure scores TM-score 0.50 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
B-cell lymphoma/leukemia 11A is a 27-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 1.65 Å, lDDT 0.77, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.59. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2024-05-24Protein Mis18-alpha✓ no confident missTM 0.57pLDDT 86FRAUD 0.66
3 structures analysed this release. AlphaFold's largest disagreement was Protein Mis18-alpha (TM 0.57) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8QXD_X at mean pLDDT 77.6, but the deposited structure scores TM-score 0.23 against that prediction — 15% of residues confidently wrong. A wrong biological context.
U4/U6.U5 small nuclear ribonucleoprotein 27 kDa protein is a 155-residue chain annotated as RNA-binding domain, RBD; the deposition resolves 61 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.23, Cα-RMSD 3.27 Å, lDDT 0.67, secondary-structure agreement Q3 50%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.36. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8VOH_B at mean pLDDT 79.3, but the deposited structure scores TM-score 0.48 against that prediction — 55% of residues confidently wrong. A wrong biological context.
Pleiotrophin is a 58-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 6.73 Å, lDDT 0.76, secondary-structure agreement Q3 79%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.31. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2024-05-10CHRNA7-FAM7A fusion protein✓ no confident missTM 0.60pLDDT 70FRAUD 0.34
2 structures analysed this release. AlphaFold's largest disagreement was CHRNA7-FAM7A fusion protein (TM 0.60) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8WXE_n at mean pLDDT 91.9, but the deposited structure scores TM-score 0.40 against that prediction — 100% of residues confidently wrong. A wrong biological context.
Signal peptide,flag tag,T cell receptor gamma variable 5,T cell receptor gamma constant 1 is a 331-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.9 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 14.29 Å, lDDT 0.30, secondary-structure agreement Q3 3%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.26. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8UIP_A at mean pLDDT 70.8, but the deposited structure scores TM-score 0.37 against that prediction — 53% of residues confidently wrong. A wrong biological context.
Ninjurin-1 is a 152-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 70.8 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 10.68 Å, lDDT 0.61, secondary-structure agreement Q3 75%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.06. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2024-04-24Sodium/calcium exchanger 1✓ no confident missTM 0.56pLDDT 82FRAUD 0.78
104 structures analysed this release. AlphaFold's largest disagreement was Sodium/calcium exchanger 1 (TM 0.56) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8Q7Q_H at mean pLDDT 81.9, but the deposited structure scores TM-score 0.42 against that prediction — 5% of residues confidently wrong. A right fold, wrong conformation/assembly.
Protein TSSC4 is a 329-residue chain; the deposition resolves 15 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 2.31 Å, lDDT 0.76, secondary-structure agreement Q3 86%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.10. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8RVE_0 at mean pLDDT 82.4, but the deposited structure scores TM-score 0.48 against that prediction — 79% of residues confidently wrong. A right fold, wrong conformation/assembly.
Vimentin is a 466-residue chain; the deposition resolves 78 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 38.27 Å, lDDT 0.73, secondary-structure agreement Q3 89%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.73. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9A3R_A at mean pLDDT 77.1, but the deposited structure scores TM-score 0.32 against that prediction — 70% of residues confidently wrong. A wrong biological context.
Vimentin is a 466-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.32, Cα-RMSD 171.92 Å, lDDT 0.82, secondary-structure agreement Q3 81%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.87. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2024-04-03DET1- and DDB1-associated protein 1Wrong biological contextTM 0.27pLDDT 76FRAUD 0.58
AlphaFold modelled 8ROY_D at mean pLDDT 76.5, but the deposited structure scores TM-score 0.27 against that prediction — 84% of residues confidently wrong. A wrong biological context.
DET1- and DDB1-associated protein 1 is a 102-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.27, Cα-RMSD 14.71 Å, lDDT 0.77, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.57. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2024-03-27Isoform Short of Insulin receptorRight fold, wrong conformation/assemblyTM 0.48pLDDT 88FRAUD 0.81
AlphaFold modelled 8VJB_A at mean pLDDT 87.7, but the deposited structure scores TM-score 0.48 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
Isoform Short of Insulin receptor is a 1370-residue chain annotated as L domain-like; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 19.51 Å, lDDT 0.85, secondary-structure agreement Q3 90%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.53. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2024-03-20Coiled-coil domain-containing protein 22Right fold, wrong conformation/assemblyTM 0.32pLDDT 89FRAUD 0.87
AlphaFold modelled 8P0V_L at mean pLDDT 88.7, but the deposited structure scores TM-score 0.32 against that prediction — 95% of residues confidently wrong. A right fold, wrong conformation/assembly.
Coiled-coil domain-containing protein 22 is a 627-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.32, Cα-RMSD 40.11 Å, lDDT 0.87, secondary-structure agreement Q3 95%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.60. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8Y7E_v at mean pLDDT 83.5, but the deposited structure scores TM-score 0.46 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
Sodium channel modifier 1 is a 230-residue chain; the deposition resolves 17 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 25.05 Å, lDDT 0.85, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.66. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2024-03-06Vacuolar protein sorting-associated protein 72 homologRight fold, wrong conformation/assemblyTM 0.33pLDDT 89FRAUD 0.73
AlphaFold modelled 8X1C_J at mean pLDDT 88.7, but the deposited structure scores TM-score 0.33 against that prediction — 94% of residues confidently wrong. A right fold, wrong conformation/assembly.
Vacuolar protein sorting-associated protein 72 homolog is a 364-residue chain; the deposition resolves 25 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.33, Cα-RMSD 17.78 Å, lDDT 0.83, secondary-structure agreement Q3 79%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.52. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8JE1_R at mean pLDDT 85.3, but the deposited structure scores TM-score 0.36 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
E3 ubiquitin-protein ligase RBX1 is a 108-residue chain; the deposition resolves 9 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 2.10 Å, lDDT 0.82, secondary-structure agreement Q3 75%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.61. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2024-02-21Small nuclear ribonucleoprotein-associated protein N peptideWrong biological contextTM 0.33pLDDT 97FRAUD 0.38
AlphaFold modelled 8TBP_E at mean pLDDT 97.0, but the deposited structure scores TM-score 0.33 against that prediction — 64% of residues confidently wrong. A wrong biological context.
Small nuclear ribonucleoprotein-associated protein N peptide is a 15-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 97.0 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.33, Cα-RMSD 6.68 Å, lDDT 0.76, secondary-structure agreement Q3 50%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.04. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2024-02-14Kinetochore-associated protein NSL1 homologRight fold, wrong conformation/assemblyTM 0.31pLDDT 89FRAUD 0.88
AlphaFold modelled 8Q5H_N at mean pLDDT 88.7, but the deposited structure scores TM-score 0.31 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
Kinetochore-associated protein NSL1 homolog is a 281-residue chain; the deposition resolves 7 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.31, Cα-RMSD 50.67 Å, lDDT 0.86, secondary-structure agreement Q3 88%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.47. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8RC4_a at mean pLDDT 82.5, but the deposited structure scores TM-score 0.34 against that prediction — 89% of residues confidently wrong. A right fold, wrong conformation/assembly.
Integrator complex subunit 1 is a 2192-residue chain; the deposition resolves 16 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.34, Cα-RMSD 48.63 Å, lDDT 0.75, secondary-structure agreement Q3 91%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.10. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2024-02-02Fibronectin✓ no confident missTM 0.43pLDDT 64FRAUD 0.57
8 structures analysed this release. AlphaFold's largest disagreement was Fibronectin (TM 0.43) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8VIS_A at mean pLDDT 82.3, but the deposited structure scores TM-score 0.25 against that prediction — 73% of residues confidently wrong. A right fold, wrong conformation/assembly.
Transmembrane protease serine 11D non-catalytic chain is a 145-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.25, Cα-RMSD 5.83 Å, lDDT 0.66, secondary-structure agreement Q3 100%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.29. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2024-01-24Microfibrillar-associated protein 1Native fold → amyloidTM 0.46pLDDT 84FRAUD 0.79
AlphaFold modelled 8QO9_K at mean pLDDT 84.1, but the deposited structure scores TM-score 0.46 against that prediction — 91% of residues confidently wrong. A native fold → amyloid.
Microfibrillar-associated protein 1 is a 439-residue chain; the deposition resolves 71 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 84.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 37.91 Å, lDDT 0.90, secondary-structure agreement Q3 92%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.31. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
2024-01-17Guanylate binding protein 1Right fold, wrong conformation/assemblyTM 0.48pLDDT 92FRAUD 0.92
AlphaFold modelled 8R1A_A at mean pLDDT 91.7, but the deposited structure scores TM-score 0.48 against that prediction — 95% of residues confidently wrong. A right fold, wrong conformation/assembly.
Guanylate binding protein 1 is a 603-residue chain annotated as GBP helical domain-like; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.7 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 66.56 Å, lDDT 0.86, secondary-structure agreement Q3 92%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.47. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8AZ6_A at mean pLDDT 74.3, but the deposited structure scores TM-score 0.22 against that prediction — 48% of residues confidently wrong. A native fold → amyloid.
Islet amyloid polypeptide is a 38-residue chain; the deposition resolves 20 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 74.3 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.22, Cα-RMSD 8.70 Å, lDDT 0.48, secondary-structure agreement Q3 45%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.49. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8IPM_C at mean pLDDT 86.1, but the deposited structure scores TM-score 0.42 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
Putative ribosome-binding factor A, mitochondrial is a 34-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 0.99 Å, lDDT 0.90, secondary-structure agreement Q3 90%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.56. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
157 structures analysed this release. AlphaFold's largest disagreement was Eukaryotic peptide chain release factor subunit 1 (TM 0.66) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8FJA_C at mean pLDDT 90.9, but the deposited structure scores TM-score 0.29 against that prediction — 30% of residues confidently wrong. A right fold, wrong conformation/assembly.
Melanoma-associated antigen 4 peptide is a 10-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.9 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 4.25 Å, lDDT 0.69, secondary-structure agreement Q3 100%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.38. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2023-12-13Transthyretin✓ no confident missTM 0.23pLDDT 98FRAUD 0.95
228 structures analysed this release. AlphaFold's largest disagreement was Transthyretin (TM 0.23) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8U5G_C at mean pLDDT 79.1, but the deposited structure scores TM-score 0.29 against that prediction — 4% of residues confidently wrong. A wrong biological context.
E3 ubiquitin-protein ligase PPP1R11 is a 46-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 2.74 Å, lDDT 0.85, secondary-structure agreement Q3 75%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.41. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2023-11-29Interleukin-18✓ no confident missTM 0.68pLDDT 90FRAUD 0.83
125 structures analysed this release. AlphaFold's largest disagreement was Interleukin-18 (TM 0.68) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8UXL_I at mean pLDDT 86.5, but the deposited structure scores TM-score 0.50 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 149-residue chain; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 11.06 Å, lDDT 0.82, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.55. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2023-11-15Guanine nucleotide-binding protein G(i) subunit alpha-1✓ no confident missTM 0.78pLDDT 94FRAUD 0.20
103 structures analysed this release. AlphaFold's largest disagreement was Guanine nucleotide-binding protein G(i) subunit alpha-1 (TM 0.78) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8IVZ_C at mean pLDDT 70.8, but the deposited structure scores TM-score 0.40 against that prediction — 33% of residues confidently wrong. A wrong biological context.
KN motif and ankyrin repeat domains 1 is a 26-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 70.8 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 4.84 Å, lDDT 0.81, secondary-structure agreement Q3 71%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.36. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2023-11-01Myosin-binding protein C, cardiac-typeRight fold, wrong conformation/assemblyTM 0.47pLDDT 85FRAUD 0.83
AlphaFold modelled 8G4L_ao at mean pLDDT 85.4, but the deposited structure scores TM-score 0.47 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
Myosin-binding protein C, cardiac-type is a 1274-residue chain; the deposition resolves 123 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 49.21 Å, lDDT 0.82, secondary-structure agreement Q3 79%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.55. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8UC6_B at mean pLDDT 88.7, but the deposited structure scores TM-score 0.49 against that prediction — 93% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calpain-7 is a 165-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 11.43 Å, lDDT 0.90, secondary-structure agreement Q3 95%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.26. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2023-10-18Retinol-binding protein 2✓ no confident missTM 0.56pLDDT 97FRAUD 0.85
254 structures analysed this release. AlphaFold's largest disagreement was Retinol-binding protein 2 (TM 0.56) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2023-10-11Sodium/calcium exchanger 1✓ no confident missTM 0.56pLDDT 82FRAUD 0.78
181 structures analysed this release. AlphaFold's largest disagreement was Sodium/calcium exchanger 1 (TM 0.56) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8BFG_A at mean pLDDT 88.0, but the deposited structure scores TM-score 0.47 against that prediction — 41% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 148-residue chain annotated as EF-hand. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 4.61 Å, lDDT 0.61, secondary-structure agreement Q3 94%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.28. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8HQY_S at mean pLDDT 72.6, but the deposited structure scores TM-score 0.14 against that prediction — 35% of residues confidently wrong. A wrong biological context.
Protein SSX2 is a 23-residue chain; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.6 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.14, Cα-RMSD 5.53 Å, lDDT 0.76, secondary-structure agreement Q3 100%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.53. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2023-09-204C1 Fab light chainRight fold, wrong conformation/assemblyTM 0.49pLDDT 97FRAUD 0.03
AlphaFold modelled 8ECZ_A at mean pLDDT 96.5, but the deposited structure scores TM-score 0.49 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
4C1 Fab light chain is a 216-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.5 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 0.68 Å, lDDT 0.96, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.78. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8TK7_D at mean pLDDT 93.7, but the deposited structure scores TM-score 0.41 against that prediction — 0% of residues confidently wrong. A wrong biological context.
Methylated-DNA--protein-cysteine methyltransferase is a 203-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 93.7 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 2.54 Å, lDDT 0.68, secondary-structure agreement Q3 30%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.52. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7Z1X_A at mean pLDDT 85.0, but the deposited structure scores TM-score 0.48 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
Gasdermin-D is a 447-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 23.16 Å, lDDT 0.84, secondary-structure agreement Q3 86%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.80. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8H9F_P at mean pLDDT 89.0, but the deposited structure scores TM-score 0.43 against that prediction — 80% of residues confidently wrong. A right fold, wrong conformation/assembly.
ATP synthase subunit ATP5MJ, mitochondrial is a 58-residue chain; the deposition resolves 19 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 7.96 Å, lDDT 0.86, secondary-structure agreement Q3 63%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.82. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8IQK_B at mean pLDDT 91.6, but the deposited structure scores TM-score 0.50 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
Bcl-2-modifying factor is a 25-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 1.68 Å, lDDT 0.84, secondary-structure agreement Q3 71%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.67. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7Y5L_A at mean pLDDT 72.5, but the deposited structure scores TM-score 0.31 against that prediction — 62% of residues confidently wrong. A wrong biological context.
Chromatin assembly factor 1 subunit A is a 273-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.5 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.31, Cα-RMSD 27.91 Å, lDDT 0.71, secondary-structure agreement Q3 87%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.41. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8IR3_z at mean pLDDT 90.8, but the deposited structure scores TM-score 0.38 against that prediction — 99% of residues confidently wrong. A right fold, wrong conformation/assembly.
Protein LLP homolog is a 129-residue chain; the deposition resolves 59 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.8 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 15.23 Å, lDDT 0.84, secondary-structure agreement Q3 87%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.28. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7Y71_A at mean pLDDT 71.3, but the deposited structure scores TM-score 0.35 against that prediction — 65% of residues confidently wrong. A wrong biological context.
Spike glycoprotein is a 1198-residue chain annotated as Immunoglobulins (2); the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 71.3 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 28.93 Å, lDDT 0.68, secondary-structure agreement Q3 76%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.83. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8HIA_A at mean pLDDT 96.6, but the deposited structure scores TM-score 0.16 against that prediction — 100% of residues confidently wrong. A native fold → amyloid.
Transforming growth factor-beta-induced protein ig-h3 is a 23-residue chain; the deposition resolves 20 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.16, Cα-RMSD 19.66 Å, lDDT 0.75, secondary-structure agreement Q3 39%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.37. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8PKP_M at mean pLDDT 77.3, but the deposited structure scores TM-score 0.23 against that prediction — 66% of residues confidently wrong. A wrong biological context.
Anaphase-promoting complex subunit 13 is a 74-residue chain annotated as TPR-like; the deposition resolves 19 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.23, Cα-RMSD 20.85 Å, lDDT 0.76, secondary-structure agreement Q3 79%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.66. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8FL0_NK at mean pLDDT 90.6, but the deposited structure scores TM-score 0.40 against that prediction — 99% of residues confidently wrong. A right fold, wrong conformation/assembly.
Protein LLP homolog is a 129-residue chain; the deposition resolves 31 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 15.10 Å, lDDT 0.89, secondary-structure agreement Q3 93%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.21. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7QI5_m at mean pLDDT 86.3, but the deposited structure scores TM-score 0.48 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
39S ribosomal protein L55, mitochondrial is a 128-residue chain; the deposition resolves 94 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 19.44 Å, lDDT 0.83, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.74. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2023-06-28Interferon-induced, double-stranded RNA-activated protein kinaseWrong biological contextTM 0.27pLDDT 76FRAUD 0.67
AlphaFold modelled 8I9J_A at mean pLDDT 75.9, but the deposited structure scores TM-score 0.27 against that prediction — 76% of residues confidently wrong. A wrong biological context.
Interferon-induced, double-stranded RNA-activated protein kinase is a 179-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 75.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.27, Cα-RMSD 17.92 Å, lDDT 0.48, secondary-structure agreement Q3 47%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.06. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2023-06-21Autophagy-related protein 13Wrong biological contextTM 0.42pLDDT 78FRAUD 0.19
AlphaFold modelled 8SQZ_E at mean pLDDT 77.5, but the deposited structure scores TM-score 0.42 against that prediction — 9% of residues confidently wrong. A wrong biological context.
Autophagy-related protein 13 is a 155-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 4.75 Å, lDDT 0.76, secondary-structure agreement Q3 58%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.36. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8OIR_Bd at mean pLDDT 86.3, but the deposited structure scores TM-score 0.48 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
39S ribosomal protein L55, mitochondrial is a 128-residue chain; the deposition resolves 94 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 19.45 Å, lDDT 0.83, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.72. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8ITY_Q at mean pLDDT 85.2, but the deposited structure scores TM-score 0.48 against that prediction — 68% of residues confidently wrong. A right fold, wrong conformation/assembly.
DNA-directed RNA polymerase III subunit RPC7 is a 223-residue chain; the deposition resolves 25 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 8.92 Å, lDDT 0.87, secondary-structure agreement Q3 90%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.59. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2023-06-01Histone H2A✓ no confident missTM 0.93pLDDT 91FRAUD 0.06
8 structures analysed this release. AlphaFold's largest disagreement was Histone H2A (TM 0.93) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8H9T_P at mean pLDDT 89.0, but the deposited structure scores TM-score 0.44 against that prediction — 83% of residues confidently wrong. A right fold, wrong conformation/assembly.
ATP synthase subunit ATP5MJ, mitochondrial is a 58-residue chain; the deposition resolves 27 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 8.14 Å, lDDT 0.86, secondary-structure agreement Q3 63%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.78. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8IUH_Q at mean pLDDT 85.2, but the deposited structure scores TM-score 0.48 against that prediction — 83% of residues confidently wrong. A right fold, wrong conformation/assembly.
DNA-directed RNA polymerase III subunit RPC7 is a 223-residue chain; the deposition resolves 25 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 9.54 Å, lDDT 0.84, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.69. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7ZNL_G at mean pLDDT 93.4, but the deposited structure scores TM-score 0.47 against that prediction — 100% of residues confidently wrong. A right fold, wrong conformation/assembly.
THO complex subunit 7 homolog is a 204-residue chain; the deposition resolves 28 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 93.4 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 14.89 Å, lDDT 0.90, secondary-structure agreement Q3 95%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.69. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 8CH6_I at mean pLDDT 85.2, but the deposited structure scores TM-score 0.37 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
Splicing factor 3A subunit 2 is a 464-residue chain; the deposition resolves 59 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 24.43 Å, lDDT 0.56, secondary-structure agreement Q3 81%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.02. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 72% novel to AlphaFold's training cutoff.
AlphaFold modelled 7ZNK_G at mean pLDDT 93.4, but the deposited structure scores TM-score 0.47 against that prediction — 100% of residues confidently wrong. A right fold, wrong conformation/assembly.
THO complex subunit 7 homolog is a 204-residue chain annotated as YVTN repeat-like/Quinoprotein amine dehydrogenase (2); the deposition resolves 32 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 93.4 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 14.89 Å, lDDT 0.90, secondary-structure agreement Q3 95%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.69. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2023-04-28alpha-synuclein✓ no confident missTM 0.24pLDDT 81FRAUD 0.79
1 structure analysed this release. AlphaFold's largest disagreement was alpha-synuclein (TM 0.24) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2023-04-26Isoform I of NeurofibrominRight fold, wrong conformation/assemblyTM 0.46pLDDT 85FRAUD 0.67
AlphaFold modelled 8E20_A at mean pLDDT 85.5, but the deposited structure scores TM-score 0.46 against that prediction — 90% of residues confidently wrong. A right fold, wrong conformation/assembly.
Isoform I of Neurofibromin is a 2818-residue chain annotated as CRAL/TRIO domain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 22.98 Å, lDDT 0.79, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.61. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7XM8_A at mean pLDDT 80.2, but the deposited structure scores TM-score 0.23 against that prediction — 55% of residues confidently wrong. A native fold → amyloid.
Glucagon is a 29-residue chain; the deposition resolves 22 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 80.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.23, Cα-RMSD 10.67 Å, lDDT 0.51, secondary-structure agreement Q3 10%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.84. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
2023-04-12NUT family member 1Wrong biological contextTM 0.40pLDDT 81FRAUD 0.17
AlphaFold modelled 7XFG_B at mean pLDDT 81.0, but the deposited structure scores TM-score 0.40 against that prediction — 25% of residues confidently wrong. A wrong biological context.
NUT family member 1 is a 16-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 3.39 Å, lDDT 0.68, secondary-structure agreement Q3 31%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.58. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7ZGK_B at mean pLDDT 79.5, but the deposited structure scores TM-score 0.44 against that prediction — 78% of residues confidently wrong. A wrong biological context.
Complement C3b alpha' chain is a 915-residue chain annotated as Terpenoid cyclases/Protein prenyltransferases (1); the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 27.54 Å, lDDT 0.76, secondary-structure agreement Q3 88%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.60. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2023-03-29Immunoglobulin heavy constant epsilonRight fold, wrong conformation/assemblyTM 0.50pLDDT 85FRAUD 0.80
AlphaFold modelled 8C1C_H at mean pLDDT 84.8, but the deposited structure scores TM-score 0.50 against that prediction — 90% of residues confidently wrong. A right fold, wrong conformation/assembly.
Immunoglobulin heavy constant epsilon is a 426-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 84.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 29.92 Å, lDDT 0.87, secondary-structure agreement Q3 89%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.70. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2023-03-27PTH1R✓ no confident missTM 0.74pLDDT 71FRAUD 0.19
12 structures analysed this release. AlphaFold's largest disagreement was PTH1R (TM 0.74) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8AR3_A at mean pLDDT 72.7, but the deposited structure scores TM-score 0.42 against that prediction — 58% of residues confidently wrong. A disorder / coiled-coil ambiguity.
Toll-like receptor 9 is a 50-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.7 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 9.99 Å, lDDT 0.72, secondary-structure agreement Q3 68%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.72. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7YCX_A at mean pLDDT 81.5, but the deposited structure scores TM-score 0.45 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
Integrator complex subunit 1 is a 2190-residue chain; the deposition resolves 34 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 56.49 Å, lDDT 0.77, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.54. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2023-03-08DET1- and DDB1-associated protein 1Wrong biological contextTM 0.26pLDDT 73FRAUD 0.23
AlphaFold modelled 8G46_E at mean pLDDT 73.0, but the deposited structure scores TM-score 0.26 against that prediction — 47% of residues confidently wrong. A wrong biological context.
DET1- and DDB1-associated protein 1 is a 106-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 73.0 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.26, Cα-RMSD 5.18 Å, lDDT 0.74, secondary-structure agreement Q3 100%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.07. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7YJ4_A at mean pLDDT 88.9, but the deposited structure scores TM-score 0.21 against that prediction — 5% of residues confidently wrong. A right fold, wrong conformation/assembly.
Insulin-like peptide INSL5 A chain is a 21-residue chain; the deposition resolves 7 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.21, Cα-RMSD 2.10 Å, lDDT 0.70, secondary-structure agreement Q3 60%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.26. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2023-02-22Intraflagellar transport protein 43 homologRight fold, wrong conformation/assemblyTM 0.37pLDDT 82FRAUD 0.48
AlphaFold modelled 8FGW_F at mean pLDDT 82.5, but the deposited structure scores TM-score 0.37 against that prediction — 70% of residues confidently wrong. A right fold, wrong conformation/assembly.
Intraflagellar transport protein 43 homolog is a 208-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 11.81 Å, lDDT 0.75, secondary-structure agreement Q3 70%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.75. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2023-02-17alpha-synuclein✓ no confident missTM 0.14pLDDT 75FRAUD 0.71
1 structure analysed this release. AlphaFold's largest disagreement was alpha-synuclein (TM 0.14) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8DWL_B at mean pLDDT 78.6, but the deposited structure scores TM-score 0.38 against that prediction — 73% of residues confidently wrong. A wrong biological context.
E3 ubiquitin-protein ligase PPP1R11 is a 46-residue chain annotated as Metallo-dependent phosphatases (3); the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 9.73 Å, lDDT 0.87, secondary-structure agreement Q3 47%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.78. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2023-02-08Epidermal growth factor receptor✓ no confident missTM 0.57pLDDT 92FRAUD 0.76
323 structures analysed this release. AlphaFold's largest disagreement was Epidermal growth factor receptor (TM 0.57) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8DTU_C at mean pLDDT 82.3, but the deposited structure scores TM-score 0.48 against that prediction — 4% of residues confidently wrong. A right fold, wrong conformation/assembly.
B-cell lymphoma/leukemia 11A is a 28-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 3.22 Å, lDDT 0.77, secondary-structure agreement Q3 75%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.20. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2023-01-25Centromere protein QRight fold, wrong conformation/assemblyTM 0.38pLDDT 89FRAUD 0.83
AlphaFold modelled 7XHN_Q at mean pLDDT 89.2, but the deposited structure scores TM-score 0.38 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
Centromere protein Q is a 274-residue chain; the deposition resolves 20 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 22.18 Å, lDDT 0.80, secondary-structure agreement Q3 87%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.64. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2023-01-19Ubiquitin✓ no confident missTM 0.92pLDDT 89FRAUD 0.07
1 structure analysed this release. AlphaFold's largest disagreement was Ubiquitin (TM 0.92) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8EB0_A at mean pLDDT 85.8, but the deposited structure scores TM-score 0.47 against that prediction — 95% of residues confidently wrong. A right fold, wrong conformation/assembly.
E3 ubiquitin-protein ligase RNF216 is a 277-residue chain annotated as Zinc/RING finger domain, C3HC4 (zinc finger) (3); the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 24.01 Å, lDDT 0.81, secondary-structure agreement Q3 95%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.49. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 76% novel to AlphaFold's training cutoff.
2023-01-11Target of rapamycin complex 2 subunit MAPKAP1✓ no confident missTM 0.35pLDDT 67FRAUD 0.66
89 structures analysed this release. AlphaFold's largest disagreement was Target of rapamycin complex 2 subunit MAPKAP1 (TM 0.35) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2023-01-04SIT1✓ no confident missTM 0.92pLDDT 93FRAUD 0.07
5 structures analysed this release. AlphaFold's largest disagreement was SIT1 (TM 0.92) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 8BFI_C at mean pLDDT 85.8, but the deposited structure scores TM-score 0.49 against that prediction — 87% of residues confidently wrong. A right fold, wrong conformation/assembly.
CCR4-NOT transcription complex subunit 11 is a 455-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 27.16 Å, lDDT 0.91, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.57. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2022-12-21Intraflagellar transport protein 43 homologRight fold, wrong conformation/assemblyTM 0.42pLDDT 84FRAUD 0.34
AlphaFold modelled 8BBE_F at mean pLDDT 84.2, but the deposited structure scores TM-score 0.42 against that prediction — 53% of residues confidently wrong. A right fold, wrong conformation/assembly.
Intraflagellar transport protein 43 homolog is a 209-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 84.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 7.01 Å, lDDT 0.81, secondary-structure agreement Q3 68%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.62. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2022-12-14Centromere protein QRight fold, wrong conformation/assemblyTM 0.39pLDDT 89FRAUD 0.83
AlphaFold modelled 7XHO_Q at mean pLDDT 89.2, but the deposited structure scores TM-score 0.39 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
Centromere protein Q is a 268-residue chain; the deposition resolves 17 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.39, Cα-RMSD 22.21 Å, lDDT 0.81, secondary-structure agreement Q3 88%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.65. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7T0Y_B at mean pLDDT 70.3, but the deposited structure scores TM-score 0.35 against that prediction — 48% of residues confidently wrong. A wrong biological context.
Ribosomal RNA processing protein 1 homolog B is a 47-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 70.3 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 11.28 Å, lDDT 0.73, secondary-structure agreement Q3 87%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.54. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2022-11-30Isoform Alpha of PaxillinRight fold, wrong conformation/assemblyTM 0.43pLDDT 93FRAUD 0.66
AlphaFold modelled 7QB0_A at mean pLDDT 93.2, but the deposited structure scores TM-score 0.43 against that prediction — 96% of residues confidently wrong. A right fold, wrong conformation/assembly.
Isoform Alpha of Paxillin is a 127-residue chain annotated as Cysteine Rich Protein (2). This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 93.2 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 12.01 Å, lDDT 0.71, secondary-structure agreement Q3 90%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.26. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8F0A_a at mean pLDDT 89.6, but the deposited structure scores TM-score 0.50 against that prediction — 96% of residues confidently wrong. A wrong biological context.
Telomeric repeat-binding factor 1 is a 27-residue chain; the deposition resolves 9 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 8.72 Å, lDDT 0.63, secondary-structure agreement Q3 33%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.58. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8EP1_E at mean pLDDT 86.6, but the deposited structure scores TM-score 0.46 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 142-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 12.79 Å, lDDT 0.75, secondary-structure agreement Q3 92%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.40. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2022-11-09Isoform Short of Insulin receptorRight fold, wrong conformation/assemblyTM 0.50pLDDT 88FRAUD 0.82
AlphaFold modelled 7YQ5_E at mean pLDDT 88.1, but the deposited structure scores TM-score 0.50 against that prediction — 93% of residues confidently wrong. A right fold, wrong conformation/assembly.
Isoform Short of Insulin receptor is a 907-residue chain annotated as L domain-like; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 18.32 Å, lDDT 0.81, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.64. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7QFB_B at mean pLDDT 77.3, but the deposited structure scores TM-score 0.21 against that prediction — 41% of residues confidently wrong. A wrong biological context.
Protein phosphatase 1 regulatory subunit 3C is a 27-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.21, Cα-RMSD 4.75 Å, lDDT 0.77, secondary-structure agreement Q3 64%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.54. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2022-10-28Postsynaptic density protein 95 (PSD95) PDZ3-SH3-GuK Module✓ no confident missTM 0.68pLDDT 86FRAUD 0.60
2 structures analysed this release. AlphaFold's largest disagreement was Postsynaptic density protein 95 (PSD95) PDZ3-SH3-GuK Module (TM 0.68) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7PGQ_F at mean pLDDT 82.5, but the deposited structure scores TM-score 0.46 against that prediction — 83% of residues confidently wrong. A right fold, wrong conformation/assembly.
Neurofibromin is a 2839-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 23.99 Å, lDDT 0.78, secondary-structure agreement Q3 86%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.70. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7WTZ_t at mean pLDDT 81.0, but the deposited structure scores TM-score 0.42 against that prediction — 81% of residues confidently wrong. A right fold, wrong conformation/assembly.
Protein LTV1 homolog is a 475-residue chain; the deposition resolves 34 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 48.70 Å, lDDT 0.80, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.49. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7VMB_C at mean pLDDT 87.1, but the deposited structure scores TM-score 0.50 against that prediction — 65% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 153-residue chain annotated as EF-hand (1); the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 9.84 Å, lDDT 0.71, secondary-structure agreement Q3 90%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.14. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2022-10-05Cilia- and flagella-associated protein 53Right fold, wrong conformation/assemblyTM 0.43pLDDT 89FRAUD 0.87
AlphaFold modelled 7UNG_3 at mean pLDDT 88.6, but the deposited structure scores TM-score 0.43 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
Cilia- and flagella-associated protein 53 is a 514-residue chain; the deposition resolves 435 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 64.65 Å, lDDT 0.93, secondary-structure agreement Q3 95%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.74. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2022-09-28Insulin-like growth factor-binding protein 5Wrong biological contextTM 0.38pLDDT 79FRAUD 0.06
AlphaFold modelled 7UFG_C at mean pLDDT 79.0, but the deposited structure scores TM-score 0.38 against that prediction — 0% of residues confidently wrong. A wrong biological context.
Insulin-like growth factor-binding protein 5 is a 272-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 1.39 Å, lDDT 0.84, secondary-structure agreement Q3 80%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.67. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7R5J_10 at mean pLDDT 81.2, but the deposited structure scores TM-score 0.43 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
Nuclear pore membrane glycoprotein 210 is a 1887-residue chain; the deposition resolves 101 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 106.09 Å, lDDT 0.90, secondary-structure agreement Q3 86%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.71. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7VF5_C at mean pLDDT 90.2, but the deposited structure scores TM-score 0.40 against that prediction — 97% of residues confidently wrong. A right fold, wrong conformation/assembly.
Pre-mRNA-splicing regulator WTAP is a 396-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.2 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 24.73 Å, lDDT 0.90, secondary-structure agreement Q3 90%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.76. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7VC9_M at mean pLDDT 75.2, but the deposited structure scores TM-score 0.36 against that prediction — 69% of residues confidently wrong. A wrong biological context.
Mitochondrial import receptor subunit TOM20 homolog is a 145-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 75.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 8.18 Å, lDDT 0.60, secondary-structure agreement Q3 71%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.55. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7SX8_D at mean pLDDT 94.1, but the deposited structure scores TM-score 0.45 against that prediction — 98% of residues confidently wrong. A right fold, wrong conformation/assembly.
Plastin-3 is a 630-residue chain annotated as Calponin-homology domain, CH-domain; the deposition resolves 7 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 94.1 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 25.32 Å, lDDT 0.69, secondary-structure agreement Q3 76%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.02. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9A1V_B at mean pLDDT 85.6, but the deposited structure scores TM-score 0.49 against that prediction — 83% of residues confidently wrong. A right fold, wrong conformation/assembly.
RPA32 is a 226-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 23.39 Å, lDDT 0.58, secondary-structure agreement Q3 70%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.63. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7PGP_F at mean pLDDT 85.9, but the deposited structure scores TM-score 0.49 against that prediction — 88% of residues confidently wrong. A wrong biological context.
Neurofibromin is a 2839-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 23.00 Å, lDDT 0.30, secondary-structure agreement Q3 52%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.23. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7V3P_A at mean pLDDT 87.7, but the deposited structure scores TM-score 0.47 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
Insulin-like growth factor 1 receptor is a 931-residue chain annotated as L domain-like; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 20.94 Å, lDDT 0.84, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.68. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
85 structures analysed this release. AlphaFold's largest disagreement was Excitatory amino acid transporter 2 (TM 0.62) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7EVO_E at mean pLDDT 85.9, but the deposited structure scores TM-score 0.50 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
RNA helicase is a 1052-residue chain annotated as RRM (RNA recognition motif) domain (3); the deposition resolves 21 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 31.57 Å, lDDT 0.78, secondary-structure agreement Q3 86%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.25. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2022-07-29PTX3 (protein complex)✓ no confident missTM 0.62pLDDT 78FRAUD 0.78
1 structure analysed this release. AlphaFold's largest disagreement was PTX3 (protein complex) (TM 0.62) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7NOZ_B at mean pLDDT 79.4, but the deposited structure scores TM-score 0.46 against that prediction — 78% of residues confidently wrong. A wrong biological context.
Complement C3 alpha chain is a 912-residue chain annotated as Terpenoid cyclases/Protein prenyltransferases (1); the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 27.65 Å, lDDT 0.79, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.63. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7OW5_C at mean pLDDT 97.1, but the deposited structure scores TM-score 0.35 against that prediction — 50% of residues confidently wrong. A wrong biological context.
KRAS peptide (VVVGAGGVGK) is a 10-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 97.1 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 4.64 Å, lDDT 0.70, secondary-structure agreement Q3 50%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.17. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7QI4_m at mean pLDDT 86.3, but the deposited structure scores TM-score 0.48 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
39S ribosomal protein L55, mitochondrial is a 128-residue chain; the deposition resolves 94 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 19.44 Å, lDDT 0.82, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.74. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 8D3C_A at mean pLDDT 86.4, but the deposited structure scores TM-score 0.16 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
von Willebrand factor is a 1469-residue chain annotated as von Willebrand factor type D domain-like; the deposition resolves 16 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.16, Cα-RMSD 19.43 Å, lDDT 0.60, secondary-structure agreement Q3 68%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.46. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7VZF_A at mean pLDDT 98.2, but the deposited structure scores TM-score 0.17 against that prediction — 100% of residues confidently wrong. A native fold → amyloid.
Superoxide dismutase [Cu-Zn] is a 154-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 98.2 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.17, Cα-RMSD 25.29 Å, lDDT 0.49, secondary-structure agreement Q3 57%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.20. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7R5K_10 at mean pLDDT 81.2, but the deposited structure scores TM-score 0.41 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
Nuclear pore membrane glycoprotein 210 is a 1887-residue chain; the deposition resolves 101 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 102.88 Å, lDDT 0.90, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.71. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2022-06-15Zinc finger CCHC domain-containing protein 8,Zinc finger CCHC domain-containing protein 8Wrong biological contextTM 0.31pLDDT 75FRAUD 0.73
AlphaFold modelled 7S7B_B at mean pLDDT 74.7, but the deposited structure scores TM-score 0.31 against that prediction — 66% of residues confidently wrong. A wrong biological context.
Zinc finger CCHC domain-containing protein 8,Zinc finger CCHC domain-containing protein 8 is a 618-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 74.7 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.31, Cα-RMSD 37.76 Å, lDDT 0.70, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.64. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2022-06-10POLY(A) BINDING PROTEIN CYTOPLASMIC 1✓ no confident missTM 0.96pLDDT 77FRAUD 0.03
2 structures analysed this release. AlphaFold's largest disagreement was POLY(A) BINDING PROTEIN CYTOPLASMIC 1 (TM 0.96) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2022-06-08Centromere protein HRight fold, wrong conformation/assemblyTM 0.40pLDDT 89FRAUD 0.78
AlphaFold modelled 7QOO_H at mean pLDDT 88.8, but the deposited structure scores TM-score 0.40 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
Centromere protein H is a 247-residue chain; the deposition resolves 15 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 19.74 Å, lDDT 0.88, secondary-structure agreement Q3 91%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.67. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2022-06-01Hypermethylated in cancer 2 protein✓ no confident missTM 0.54pLDDT 83FRAUD 0.36
88 structures analysed this release. AlphaFold's largest disagreement was Hypermethylated in cancer 2 protein (TM 0.54) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7WPS_A at mean pLDDT 86.3, but the deposited structure scores TM-score 0.28 against that prediction — 85% of residues confidently wrong. A right fold, wrong conformation/assembly.
von Willebrand antigen 2 is a 741-residue chain; the deposition resolves 28 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.28, Cα-RMSD 19.29 Å, lDDT 0.59, secondary-structure agreement Q3 67%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.31. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2022-05-18Centromere protein HRight fold, wrong conformation/assemblyTM 0.39pLDDT 91FRAUD 0.85
AlphaFold modelled 7YYH_H at mean pLDDT 90.9, but the deposited structure scores TM-score 0.39 against that prediction — 98% of residues confidently wrong. A right fold, wrong conformation/assembly.
Centromere protein H is a 247-residue chain; the deposition resolves 23 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.9 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.39, Cα-RMSD 20.02 Å, lDDT 0.92, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.34. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
194 structures analysed this release. AlphaFold's largest disagreement was Lymphocyte activation gene 3 protein (TM 0.57) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7W3J_e at mean pLDDT 73.7, but the deposited structure scores TM-score 0.32 against that prediction — 52% of residues confidently wrong. A wrong biological context.
26S proteasome complex subunit DSS1 is a 70-residue chain; the deposition resolves 48 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 73.7 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.32, Cα-RMSD 9.07 Å, lDDT 0.69, secondary-structure agreement Q3 62%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.40. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2022-04-27Centromere protein HRight fold, wrong conformation/assemblyTM 0.47pLDDT 93FRAUD 0.83
AlphaFold modelled 7R5V_H at mean pLDDT 93.3, but the deposited structure scores TM-score 0.47 against that prediction — 99% of residues confidently wrong. A right fold, wrong conformation/assembly.
Centromere protein H is a 247-residue chain; the deposition resolves 13 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 93.3 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 19.82 Å, lDDT 0.96, secondary-structure agreement Q3 94%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.53. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7WJI_B at mean pLDDT 79.9, but the deposited structure scores TM-score 0.42 against that prediction — 86% of residues confidently wrong. A wrong biological context.
Protein unc-79 homolog is a 2658-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 72.95 Å, lDDT 0.70, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.38. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7AKK_A at mean pLDDT 79.2, but the deposited structure scores TM-score 0.41 against that prediction — 77% of residues confidently wrong. A wrong biological context.
Complement C3b alpha' chain is a 898-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 30.89 Å, lDDT 0.73, secondary-structure agreement Q3 81%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.56. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7SHQ_B at mean pLDDT 86.2, but the deposited structure scores TM-score 0.43 against that prediction — 74% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 148-residue chain annotated as EF-hand (1); the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 8.53 Å, lDDT 0.78, secondary-structure agreement Q3 94%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.44. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7TV9_B at mean pLDDT 79.5, but the deposited structure scores TM-score 0.45 against that prediction — 78% of residues confidently wrong. A wrong biological context.
Complement C3b alpha' chain is a 915-residue chain annotated as Terpenoid cyclases/Protein prenyltransferases; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 27.64 Å, lDDT 0.78, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.61. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7FB5_B at mean pLDDT 72.7, but the deposited structure scores TM-score 0.34 against that prediction — 0% of residues confidently wrong. A wrong biological context.
Reticulophagy regulator 1 is a 26-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.7 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.34, Cα-RMSD 3.28 Å, lDDT 0.78, secondary-structure agreement Q3 62%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.26. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7PFP_A at mean pLDDT 87.4, but the deposited structure scores TM-score 0.38 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
Uromodulin is a 640-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 32.96 Å, lDDT 0.79, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.43. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2022-03-09Transmembrane protein 106B✓ no confident missTM 0.18pLDDT 94FRAUD 0.88
81 structures analysed this release. AlphaFold's largest disagreement was Transmembrane protein 106B (TM 0.18) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2022-03-02Epidermal growth factor receptor✓ no confident missTM 0.53pLDDT 92FRAUD 0.84
254 structures analysed this release. AlphaFold's largest disagreement was Epidermal growth factor receptor (TM 0.53) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7PNF_B at mean pLDDT 86.4, but the deposited structure scores TM-score 0.20 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
von Willebrand factor is a 1247-residue chain annotated as von Willebrand factor type D domain-like; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.20, Cα-RMSD 19.39 Å, lDDT 0.58, secondary-structure agreement Q3 71%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.43. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7VBC_I at mean pLDDT 85.5, but the deposited structure scores TM-score 0.40 against that prediction — 92% of residues confidently wrong. A wrong biological context.
DNA-directed RNA polymerase I subunit RPA12 is a 126-residue chain annotated as RNA polymerase subunits; the deposition resolves 16 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 18.00 Å, lDDT 0.54, secondary-structure agreement Q3 71%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.19. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7K7U_A at mean pLDDT 94.1, but the deposited structure scores TM-score 0.49 against that prediction — 90% of residues confidently wrong. A right fold, wrong conformation/assembly.
Beta-crystallin B2 is a 205-residue chain annotated as gamma-Crystallin-like; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 94.1 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 17.43 Å, lDDT 0.92, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.87. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2022-02-02Isoform Short of Insulin receptorRight fold, wrong conformation/assemblyTM 0.50pLDDT 88FRAUD 0.84
AlphaFold modelled 7PG3_A at mean pLDDT 88.4, but the deposited structure scores TM-score 0.50 against that prediction — 94% of residues confidently wrong. A right fold, wrong conformation/assembly.
Isoform Short of Insulin receptor is a 1382-residue chain annotated as Growth factor receptor domain-like; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 19.49 Å, lDDT 0.79, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.59. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2022-01-26Serine/threonine-protein kinase B-raf✓ no confident missTM 0.63pLDDT 84FRAUD 0.78
215 structures analysed this release. AlphaFold's largest disagreement was Serine/threonine-protein kinase B-raf (TM 0.63) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7LF4_D at mean pLDDT 87.5, but the deposited structure scores TM-score 0.29 against that prediction — 50% of residues confidently wrong. A wrong biological context.
Transcription factor p65 is a 22-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 4.43 Å, lDDT 0.56, secondary-structure agreement Q3 0%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.49. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7BAG_B at mean pLDDT 79.1, but the deposited structure scores TM-score 0.45 against that prediction — 77% of residues confidently wrong. A wrong biological context.
Complement C3 is a 915-residue chain annotated as Terpenoid cyclases/Protein prenyltransferases; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 27.30 Å, lDDT 0.80, secondary-structure agreement Q3 86%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.65. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2022-01-05Guanine nucleotide-binding protein G(i) subunit alpha-1✓ no confident missTM 0.71pLDDT 94FRAUD 0.27
61 structures analysed this release. AlphaFold's largest disagreement was Guanine nucleotide-binding protein G(i) subunit alpha-1 (TM 0.71) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7SX3_D at mean pLDDT 81.0, but the deposited structure scores TM-score 0.42 against that prediction — 89% of residues confidently wrong. A right fold, wrong conformation/assembly.
UNC79,Protein unc-79 homolog,Protein unc-79 homolog is a 2561-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.42, Cα-RMSD 62.62 Å, lDDT 0.75, secondary-structure agreement Q3 87%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.55. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2021-12-22Epidermal growth factor receptor✓ no confident missTM 0.53pLDDT 92FRAUD 0.84
91 structures analysed this release. AlphaFold's largest disagreement was Epidermal growth factor receptor (TM 0.53) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2021-12-15Large proline-rich protein BAG6Wrong biological contextTM 0.49pLDDT 76FRAUD 0.30
AlphaFold modelled 7RUA_D at mean pLDDT 76.0, but the deposited structure scores TM-score 0.49 against that prediction — 49% of residues confidently wrong. A wrong biological context.
Large proline-rich protein BAG6 is a 132-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 6.54 Å, lDDT 0.81, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.35. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7PKS_a at mean pLDDT 82.8, but the deposited structure scores TM-score 0.43 against that prediction — 89% of residues confidently wrong. A right fold, wrong conformation/assembly.
Integrator complex subunit 1 is a 2190-residue chain; the deposition resolves 33 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 53.32 Å, lDDT 0.76, secondary-structure agreement Q3 86%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.15. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7VNP_B at mean pLDDT 86.5, but the deposited structure scores TM-score 0.43 against that prediction — 85% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-3 is a 149-residue chain annotated as EF-hand (1); the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 11.82 Å, lDDT 0.62, secondary-structure agreement Q3 83%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.04. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2021-11-24Tyrosine-protein kinase SYK✓ no confident missTM 0.56pLDDT 91FRAUD 0.58
65 structures analysed this release. AlphaFold's largest disagreement was Tyrosine-protein kinase SYK (TM 0.56) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7PGU_F at mean pLDDT 83.9, but the deposited structure scores TM-score 0.43 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
Neurofibromin is a 2839-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 24.05 Å, lDDT 0.80, secondary-structure agreement Q3 88%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.72. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7PFO_Q at mean pLDDT 70.7, but the deposited structure scores TM-score 0.35 against that prediction — 58% of residues confidently wrong. A wrong biological context.
Claspin is a 1371-residue chain; the deposition resolves 21 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 70.7 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 21.51 Å, lDDT 0.67, secondary-structure agreement Q3 47%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.86. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2021-11-03Double homeobox protein 4-like protein 2Right fold, wrong conformation/assemblyTM 0.49pLDDT 96FRAUD 0.93
AlphaFold modelled 7DW5_A at mean pLDDT 95.6, but the deposited structure scores TM-score 0.49 against that prediction — 100% of residues confidently wrong. A right fold, wrong conformation/assembly.
Double homeobox protein 4-like protein 2 is a 150-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 95.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 21.97 Å, lDDT 0.96, secondary-structure agreement Q3 95%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.57. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7FJI_Q at mean pLDDT 84.3, but the deposited structure scores TM-score 0.35 against that prediction — 95% of residues confidently wrong. A right fold, wrong conformation/assembly.
DNA-directed RNA polymerase III subunit RPC7 is a 223-residue chain; the deposition resolves 20 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 84.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 17.60 Å, lDDT 0.57, secondary-structure agreement Q3 71%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.08. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
129 structures analysed this release. AlphaFold's largest disagreement was Peptidyl-prolyl cis-trans isomerase NIMA-interacting 1 (TM 0.61) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7OPC_V at mean pLDDT 83.7, but the deposited structure scores TM-score 0.26 against that prediction — 94% of residues confidently wrong. A wrong biological context.
RNA polymerase II-associated factor 1 homolog is a 531-residue chain; the deposition resolves 28 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.26, Cα-RMSD 40.64 Å, lDDT 0.54, secondary-structure agreement Q3 63%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.05. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7OPD_V at mean pLDDT 83.7, but the deposited structure scores TM-score 0.26 against that prediction — 94% of residues confidently wrong. A wrong biological context.
RNA polymerase II-associated factor 1 homolog is a 531-residue chain; the deposition resolves 28 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.26, Cα-RMSD 40.64 Å, lDDT 0.54, secondary-structure agreement Q3 63%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.05. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2021-09-29Isoform 3 of Nuclear factor erythroid 2-related factor 2✓ no confident missTM 0.62pLDDT 87FRAUD 0.32
82 structures analysed this release. AlphaFold's largest disagreement was Isoform 3 of Nuclear factor erythroid 2-related factor 2 (TM 0.62) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2021-09-22Active regulator of SIRT1Right fold, wrong conformation/assemblyTM 0.25pLDDT 82FRAUD 0.81
AlphaFold modelled 7MQA_LY at mean pLDDT 81.7, but the deposited structure scores TM-score 0.25 against that prediction — 87% of residues confidently wrong. A right fold, wrong conformation/assembly.
Active regulator of SIRT1 is a 136-residue chain; the deposition resolves 72 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.25, Cα-RMSD 33.12 Å, lDDT 0.83, secondary-structure agreement Q3 88%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.55. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7PE7_G at mean pLDDT 72.0, but the deposited structure scores TM-score 0.37 against that prediction — 64% of residues confidently wrong. A wrong biological context.
Target of rapamycin complex 2 subunit MAPKAP1 is a 522-residue chain; the deposition resolves 10 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.0 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 24.35 Å, lDDT 0.80, secondary-structure agreement Q3 58%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.41. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7PE9_G at mean pLDDT 72.0, but the deposited structure scores TM-score 0.37 against that prediction — 64% of residues confidently wrong. A wrong biological context.
Target of rapamycin complex 2 subunit MAPKAP1 is a 522-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.0 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 24.34 Å, lDDT 0.80, secondary-structure agreement Q3 58%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.38. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2021-09-01Akirin-2✓ no confident missTM 0.71pLDDT 66FRAUD 0.19
104 structures analysed this release. AlphaFold's largest disagreement was Akirin-2 (TM 0.71) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7A1S_B at mean pLDDT 97.4, but the deposited structure scores TM-score 0.43 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
TANKYRASE-2 is a 20-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 97.4 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 6.64 Å, lDDT 0.77, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.30. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7NFE_J at mean pLDDT 85.7, but the deposited structure scores TM-score 0.49 against that prediction — 84% of residues confidently wrong. A right fold, wrong conformation/assembly.
DNA ligase 4 is a 911-residue chain; the deposition resolves 10 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 10.46 Å, lDDT 0.74, secondary-structure agreement Q3 80%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.24. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
77 structures analysed this release. AlphaFold's largest disagreement was Guanylate cyclase soluble subunit beta-1 (TM 0.54) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7NJ1_B at mean pLDDT 79.9, but the deposited structure scores TM-score 0.24 against that prediction — 82% of residues confidently wrong. A wrong biological context.
Securin is a 202-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.24, Cα-RMSD 25.83 Å, lDDT 0.79, secondary-structure agreement Q3 73%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.78. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2021-07-28Gene for histone H3 (germline gene)Whole-fold mismatchTM 0.37pLDDT 79FRAUD 0.50
AlphaFold modelled 7D87_E at mean pLDDT 79.4, but the deposited structure scores TM-score 0.37 against that prediction. A whole-fold mismatch.
Gene for histone H3 (germline gene) is a protein chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD —, lDDT 0.00.
Confidence is only weakly informative. Mechanistically, secondary structure and tertiary packing both diverge from the prediction — AlphaFold committed to the wrong fold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7EB9_A at mean pLDDT 90.1, but the deposited structure scores TM-score 0.45 against that prediction — 97% of residues confidently wrong. A right fold, wrong conformation/assembly.
tetra ubiquitin is a 304-residue chain annotated as Phosphatidylinositol 3-kinase Catalytic Subunit; Chain A, domain 1 (3); the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.1 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 24.32 Å, lDDT 0.91, secondary-structure agreement Q3 94%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.69. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7CJB_D at mean pLDDT 92.9, but the deposited structure scores TM-score 0.39 against that prediction — 30% of residues confidently wrong. A right fold, wrong conformation/assembly.
THR-LEU-TYR-TYR-MET-ALA-PRO-GLU-HIS-LEU-ASN-ASP-VAL-ASN-ALA is a 15-residue chain; the deposition resolves 16 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 92.9 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.39, Cα-RMSD 4.06 Å, lDDT 0.61, secondary-structure agreement Q3 100%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.61. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7D0O_B at mean pLDDT 71.3, but the deposited structure scores TM-score 0.50 against that prediction — 0% of residues confidently wrong. A wrong biological context.
BRD1 protein is a 50-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 71.3 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 2.92 Å, lDDT 0.83, secondary-structure agreement Q3 25%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.54. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6ZFV_A at mean pLDDT 77.6, but the deposited structure scores TM-score 0.34 against that prediction — 65% of residues confidently wrong. A disorder / coiled-coil ambiguity.
Endothelial transcription factor GATA-2 is a 63-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.34, Cα-RMSD 22.77 Å, lDDT 0.35, secondary-structure agreement Q3 79%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.81. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7ETW_B at mean pLDDT 78.4, but the deposited structure scores TM-score 0.50 against that prediction — 78% of residues confidently wrong. A wrong biological context.
Sterol regulatory element-binding protein cleavage-activating protein is a 735-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 18.80 Å, lDDT 0.65, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.53. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 75% novel to AlphaFold's training cutoff.
AlphaFold modelled 7AXN_P at mean pLDDT 91.6, but the deposited structure scores TM-score 0.33 against that prediction. A whole-fold mismatch.
Peptidyl-prolyl cis-trans isomerase NIMA-interacting 1 is a protein chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.33, Cα-RMSD —, lDDT 0.00.
Confidence is only weakly informative. Mechanistically, secondary structure and tertiary packing both diverge from the prediction — AlphaFold committed to the wrong fold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7MO7_A at mean pLDDT 87.8, but the deposited structure scores TM-score 0.49 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
Hepatocyte growth factor is a 728-residue chain annotated as Trypsin-like serine proteases; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 17.96 Å, lDDT 0.80, secondary-structure agreement Q3 89%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.60. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 9A0U_G at mean pLDDT 98.3, but the deposited structure scores TM-score 0.30 against that prediction — 93% of residues confidently wrong. A native fold → amyloid.
Alpha-enolase is a 14-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 98.3 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.30, Cα-RMSD 8.88 Å, lDDT 0.52, secondary-structure agreement Q3 29%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.46. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7NFX_z at mean pLDDT 74.5, but the deposited structure scores TM-score 0.29 against that prediction — 0% of residues confidently wrong. A wrong biological context.
Signal recognition particle subunit SRP72 is a 671-residue chain; the deposition resolves 55 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 74.5 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 2.95 Å, lDDT 0.74, secondary-structure agreement Q3 100%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.57. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2021-05-26Mediator of RNA polymerase II transcription subunit 19Right fold, wrong conformation/assemblyTM 0.49pLDDT 81FRAUD 0.35
AlphaFold modelled 7ENA_s at mean pLDDT 81.0, but the deposited structure scores TM-score 0.49 against that prediction — 58% of residues confidently wrong. A right fold, wrong conformation/assembly.
Mediator of RNA polymerase II transcription subunit 19 is a 244-residue chain; the deposition resolves 76 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 7.52 Å, lDDT 0.62, secondary-structure agreement Q3 69%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.53. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7C41_D at mean pLDDT 95.7, but the deposited structure scores TM-score 0.41 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
HRAS-like suppressor 3 is a 10-residue chain annotated as P-loop containing nucleotide triphosphate hydrolases (3); the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 95.7 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 0.73 Å, lDDT 0.87, secondary-structure agreement Q3 70%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.35. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7EGD_H at mean pLDDT 87.9, but the deposited structure scores TM-score 0.49 against that prediction — 97% of residues confidently wrong. A right fold, wrong conformation/assembly.
Transcription initiation factor TFIID subunit 8 is a 310-residue chain; the deposition resolves 24 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 27.18 Å, lDDT 0.86, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.81. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2021-05-06alpha-synuclein✓ no confident missTM 0.16pLDDT 75FRAUD 0.73
1 structure analysed this release. AlphaFold's largest disagreement was alpha-synuclein (TM 0.16) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2021-05-05Mediator of RNA polymerase II transcription subunit 14Right fold, wrong conformation/assemblyTM 0.44pLDDT 84FRAUD 0.82
AlphaFold modelled 7NVR_l at mean pLDDT 83.5, but the deposited structure scores TM-score 0.44 against that prediction — 93% of residues confidently wrong. A right fold, wrong conformation/assembly.
Mediator of RNA polymerase II transcription subunit 14 is a 1454-residue chain; the deposition resolves 58 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 36.98 Å, lDDT 0.72, secondary-structure agreement Q3 79%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.36. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7CP9_C at mean pLDDT 77.6, but the deposited structure scores TM-score 0.46 against that prediction — 55% of residues confidently wrong. A wrong biological context.
Mitochondrial import receptor subunit TOM6 homolog is a 74-residue chain; the deposition resolves 10 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 6.05 Å, lDDT 0.77, secondary-structure agreement Q3 78%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.87. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 9A18_A at mean pLDDT 84.9, but the deposited structure scores TM-score 0.50 against that prediction — 82% of residues confidently wrong. A right fold, wrong conformation/assembly.
RAGE is a 383-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 84.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 43.51 Å, lDDT 0.78, secondary-structure agreement Q3 79%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.72. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2021-04-21Activator of 90 kDa heat shock protein ATPase homolog 1✓ no confident missTM 0.43pLDDT 83FRAUD 0.82
99 structures analysed this release. AlphaFold's largest disagreement was Activator of 90 kDa heat shock protein ATPase homolog 1 (TM 0.43) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7LSY_X at mean pLDDT 85.8, but the deposited structure scores TM-score 0.47 against that prediction — 84% of residues confidently wrong. A right fold, wrong conformation/assembly.
DNA ligase 4 is a 911-residue chain; the deposition resolves 17 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 11.99 Å, lDDT 0.80, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.42. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2021-04-07E3 ubiquitin-protein ligase Mdm2✓ no confident missTM 0.50pLDDT 90FRAUD 0.77
90 structures analysed this release. AlphaFold's largest disagreement was E3 ubiquitin-protein ligase Mdm2 (TM 0.50) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7DVQ_v at mean pLDDT 83.5, but the deposited structure scores TM-score 0.46 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
Sodium channel modifier 1 is a 230-residue chain; the deposition resolves 49 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 25.05 Å, lDDT 0.85, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.66. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2021-03-24Mediator of RNA polymerase II transcription subunit 9Right fold, wrong conformation/assemblyTM 0.41pLDDT 95FRAUD 0.48
AlphaFold modelled 7LBM_u at mean pLDDT 95.1, but the deposited structure scores TM-score 0.41 against that prediction — 98% of residues confidently wrong. A right fold, wrong conformation/assembly.
Mediator of RNA polymerase II transcription subunit 9 is a 146-residue chain; the deposition resolves 56 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 95.1 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 7.84 Å, lDDT 0.83, secondary-structure agreement Q3 90%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.02. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7DN3_Q at mean pLDDT 84.3, but the deposited structure scores TM-score 0.35 against that prediction — 95% of residues confidently wrong. A right fold, wrong conformation/assembly.
DNA-directed RNA polymerase III subunit RPC7 is a 223-residue chain; the deposition resolves 20 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 84.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 17.60 Å, lDDT 0.57, secondary-structure agreement Q3 71%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.08. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2021-03-10Fanconi anemia group B proteinWrong biological contextTM 0.47pLDDT 78FRAUD 0.74
AlphaFold modelled 7KZV_B at mean pLDDT 78.0, but the deposited structure scores TM-score 0.47 against that prediction — 77% of residues confidently wrong. A wrong biological context.
Fanconi anemia group B protein is a 884-residue chain; the deposition resolves 19 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 31.78 Å, lDDT 0.81, secondary-structure agreement Q3 80%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.75. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 7BL1_AAA at mean pLDDT 88.0, but the deposited structure scores TM-score 0.40 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
UV radiation resistance-associated gene protein is a 699-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 18.40 Å, lDDT 0.58, secondary-structure agreement Q3 80%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.59. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6Z1O_A at mean pLDDT 96.0, but the deposited structure scores TM-score 0.28 against that prediction — 97% of residues confidently wrong. A native fold → amyloid.
lambda 3 immunoglobulin light chain fragment, residues 2-116 is a 89-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.0 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.28, Cα-RMSD 20.21 Å, lDDT 0.54, secondary-structure agreement Q3 56%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.35. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7D59_Q at mean pLDDT 78.9, but the deposited structure scores TM-score 0.36 against that prediction — 77% of residues confidently wrong. A wrong biological context.
DNA-directed RNA polymerase III subunit RPC7 is a 223-residue chain; the deposition resolves 17 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 24.07 Å, lDDT 0.75, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.39. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
4 structures analysed this release. AlphaFold's largest disagreement was DNA polymerase iota peptide 573-584 (TM 0.39) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7ABI_F at mean pLDDT 89.9, but the deposited structure scores TM-score 0.46 against that prediction — 99% of residues confidently wrong. A right fold, wrong conformation/assembly.
Splicing factor 3A subunit 2 is a 464-residue chain annotated as Sm-like ribonucleoproteins; the deposition resolves 56 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 19.54 Å, lDDT 0.68, secondary-structure agreement Q3 72%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.26. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 72% novel to AlphaFold's training cutoff.
AlphaFold modelled 7AEA_Q at mean pLDDT 84.0, but the deposited structure scores TM-score 0.41 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
DNA-directed RNA polymerase III subunit RPC7 is a 223-residue chain; the deposition resolves 20 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 84.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 12.62 Å, lDDT 0.80, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.55. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
82 structures analysed this release. AlphaFold's largest disagreement was Transcription initiation factor TFIID subunit 1 (TM 0.58) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2021-01-20CD81 antigen✓ no confident missTM 0.57pLDDT 88FRAUD 0.45
113 structures analysed this release. AlphaFold's largest disagreement was CD81 antigen (TM 0.57) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6VE5_B at mean pLDDT 83.2, but the deposited structure scores TM-score 0.45 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
Shieldin complex subunit 3 is a 35-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 2.54 Å, lDDT 0.92, secondary-structure agreement Q3 63%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.73. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2021-01-06Guanine nucleotide-binding protein G(i) subunit alpha-1✓ no confident missTM 0.57pLDDT 94FRAUD 0.71
69 structures analysed this release. AlphaFold's largest disagreement was Guanine nucleotide-binding protein G(i) subunit alpha-1 (TM 0.57) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
54 structures analysed this release. AlphaFold's largest disagreement was General transcription factor IIH subunit 1 (TM 0.62) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
583 structures analysed this release. AlphaFold's largest disagreement was Splicing factor 3A subunit 3 (TM 0.60) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 9A0Q_A at mean pLDDT 73.7, but the deposited structure scores TM-score 0.49 against that prediction — 44% of residues confidently wrong. A disorder / coiled-coil ambiguity.
Transcriptional enhancer factor TEF-1 is a 87-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 73.7 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 12.89 Å, lDDT 0.61, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.60. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7DL2_C at mean pLDDT 90.4, but the deposited structure scores TM-score 0.44 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
Hamartin is a 1164-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 90.4 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 79.46 Å, lDDT 0.94, secondary-structure agreement Q3 96%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.78. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7JPU_A at mean pLDDT 80.3, but the deposited structure scores TM-score 0.41 against that prediction — 83% of residues confidently wrong. A right fold, wrong conformation/assembly.
Lymphocyte antigen 75 is a 1722-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 80.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 24.55 Å, lDDT 0.67, secondary-structure agreement Q3 74%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.57. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2020-12-04SPINDOC✓ no confident missTM 0.20pLDDT 52FRAUD 0.50
2 structures analysed this release. AlphaFold's largest disagreement was SPINDOC (TM 0.20) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 7ADO_D at mean pLDDT 70.2, but the deposited structure scores TM-score 0.37 against that prediction — 47% of residues confidently wrong. A wrong biological context.
ER membrane protein complex subunit 4 is a 183-residue chain; the deposition resolves 10 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 70.2 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 23.11 Å, lDDT 0.77, secondary-structure agreement Q3 58%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.24. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2020-11-25Integrator complex subunit 8✓ no confident missTM 0.64pLDDT 85FRAUD 0.66
154 structures analysed this release. AlphaFold's largest disagreement was Integrator complex subunit 8 (TM 0.64) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6ZWO_H at mean pLDDT 72.0, but the deposited structure scores TM-score 0.39 against that prediction — 64% of residues confidently wrong. A wrong biological context.
Target of rapamycin complex 2 subunit MAPKAP1 is a 521-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.0 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.39, Cα-RMSD 23.97 Å, lDDT 0.80, secondary-structure agreement Q3 68%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.47. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2020-11-11Neurosecretory protein VGFDisorder / coiled-coil ambiguityTM 0.37pLDDT 82FRAUD 0.46
AlphaFold modelled 7D16_A at mean pLDDT 82.1, but the deposited structure scores TM-score 0.37 against that prediction — 67% of residues confidently wrong. A disorder / coiled-coil ambiguity.
Neurosecretory protein VGF is a 39-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 10.40 Å, lDDT 0.60, secondary-structure agreement Q3 54%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.29. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6TQL_A at mean pLDDT 86.3, but the deposited structure scores TM-score 0.45 against that prediction — 89% of residues confidently wrong. A right fold, wrong conformation/assembly.
Uromodulin is a 296-residue chain annotated as Zona pellucida, ZP-C domain (2); the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 26.61 Å, lDDT 0.83, secondary-structure agreement Q3 81%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.50. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6WX9_B at mean pLDDT 85.6, but the deposited structure scores TM-score 0.26 against that prediction — 85% of residues confidently wrong. A native fold → amyloid.
Transcription factor SOX-2 is a 90-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.26, Cα-RMSD 9.96 Å, lDDT 0.83, secondary-structure agreement Q3 38%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.11. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
2020-10-211-phosphatidylinositol 3-phosphate 5-kinase✓ no confident missTM 0.59pLDDT 70FRAUD 0.52
150 structures analysed this release. AlphaFold's largest disagreement was 1-phosphatidylinositol 3-phosphate 5-kinase (TM 0.59) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 9A0K_2 at mean pLDDT 83.8, but the deposited structure scores TM-score 0.35 against that prediction — 83% of residues confidently wrong. A right fold, wrong conformation/assembly.
SMARCC2 is a 1214-residue chain; the deposition resolves 20 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 31.08 Å, lDDT 0.74, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.81. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7A5P_G at mean pLDDT 88.9, but the deposited structure scores TM-score 0.43 against that prediction — 95% of residues confidently wrong. A right fold, wrong conformation/assembly.
Pre-mRNA-processing factor 19 is a 504-residue chain; the deposition resolves 48 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 19.32 Å, lDDT 0.84, secondary-structure agreement Q3 94%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.67. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2020-10-08KCNQ1 channel-forming domain✓ no confident missTM 0.76pLDDT 68FRAUD 0.16
2 structures analysed this release. AlphaFold's largest disagreement was KCNQ1 channel-forming domain (TM 0.76) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6TMO_C at mean pLDDT 89.3, but the deposited structure scores TM-score 0.25 against that prediction — 40% of residues confidently wrong. A wrong biological context.
EAAGIGILTV is a 10-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.25, Cα-RMSD 4.09 Å, lDDT 0.44, secondary-structure agreement Q3 0%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.12. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6ZRQ_A at mean pLDDT 71.3, but the deposited structure scores TM-score 0.15 against that prediction — 39% of residues confidently wrong. A native fold → amyloid.
Islet amyloid polypeptide is a 37-residue chain; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 71.3 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.15, Cα-RMSD 7.41 Å, lDDT 0.37, secondary-structure agreement Q3 78%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.70. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6ZMW_z at mean pLDDT 76.4, but the deposited structure scores TM-score 0.31 against that prediction — 68% of residues confidently wrong. A wrong biological context.
Eukaryotic translation initiation factor 3 subunit J is a 258-residue chain; the deposition resolves 57 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.31, Cα-RMSD 29.04 Å, lDDT 0.35, secondary-structure agreement Q3 76%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.01. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6YBW_z at mean pLDDT 76.4, but the deposited structure scores TM-score 0.31 against that prediction — 68% of residues confidently wrong. A wrong biological context.
Eukaryotic translation initiation factor 3 subunit J is a 258-residue chain; the deposition resolves 26 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.31, Cα-RMSD 28.94 Å, lDDT 0.35, secondary-structure agreement Q3 75%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.02. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2020-09-09PC4 and SFRS1-interacting proteinRight fold, wrong conformation/assemblyTM 0.47pLDDT 86FRAUD 0.81
AlphaFold modelled 6TVM_A at mean pLDDT 85.7, but the deposited structure scores TM-score 0.47 against that prediction — 78% of residues confidently wrong. A right fold, wrong conformation/assembly.
PC4 and SFRS1-interacting protein is a 128-residue chain annotated as HIV integrase-binding domain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 24.97 Å, lDDT 0.71, secondary-structure agreement Q3 78%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.63. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 7BV4_C at mean pLDDT 94.5, but the deposited structure scores TM-score 0.38 against that prediction — 27% of residues confidently wrong. A wrong biological context.
Syntaxin-17 is a 22-residue chain annotated as Ubiquitin-like; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 94.5 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 4.12 Å, lDDT 0.57, secondary-structure agreement Q3 0%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.60. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6LSS_z at mean pLDDT 91.7, but the deposited structure scores TM-score 0.50 against that prediction — 24% of residues confidently wrong. A right fold, wrong conformation/assembly.
Protein LLP homolog is a 129-residue chain; the deposition resolves 48 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.7 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 4.08 Å, lDDT 0.86, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.35. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6Z32_A at mean pLDDT 78.1, but the deposited structure scores TM-score 0.40 against that prediction — 80% of residues confidently wrong. A wrong biological context.
Cation-independent mannose-6-phosphate receptor is a 737-residue chain annotated as Mannose 6-phosphate receptor domain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 30.83 Å, lDDT 0.85, secondary-structure agreement Q3 86%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.58. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2020-08-12Coiled-coil domain-containing protein 124Right fold, wrong conformation/assemblyTM 0.50pLDDT 89FRAUD 0.70
AlphaFold modelled 6ZME_CE at mean pLDDT 88.7, but the deposited structure scores TM-score 0.50 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
Coiled-coil domain-containing protein 124 is a 223-residue chain; the deposition resolves 88 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 19.58 Å, lDDT 0.95, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.83. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2020-08-05Oxytocin receptor✓ no confident missTM 0.56pLDDT 87FRAUD 0.60
301 structures analysed this release. AlphaFold's largest disagreement was Oxytocin receptor (TM 0.56) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2020-07-29Coiled-coil domain-containing protein 124Right fold, wrong conformation/assemblyTM 0.45pLDDT 88FRAUD 0.74
AlphaFold modelled 6ZM7_CE at mean pLDDT 88.4, but the deposited structure scores TM-score 0.45 against that prediction — 94% of residues confidently wrong. A right fold, wrong conformation/assembly.
Coiled-coil domain-containing protein 124 is a 223-residue chain; the deposition resolves 86 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 23.26 Å, lDDT 0.93, secondary-structure agreement Q3 94%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.83. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6VSK_A at mean pLDDT 81.1, but the deposited structure scores TM-score 0.49 against that prediction — 87% of residues confidently wrong. A right fold, wrong conformation/assembly.
Phosphatidylinositol 3,4,5-trisphosphate-dependent Rac exchanger 1 protein is a 94-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 21.63 Å, lDDT 0.85, secondary-structure agreement Q3 80%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.14. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 71% novel to AlphaFold's training cutoff.
122 structures analysed this release. AlphaFold's largest disagreement was Sarcoplasmic/endoplasmic reticulum calcium ATPase 2 (TM 0.69) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6U4Y_A at mean pLDDT 89.4, but the deposited structure scores TM-score 0.48 against that prediction — 97% of residues confidently wrong. A right fold, wrong conformation/assembly.
Histone-lysine N-methyltransferase EZH2 is a 224-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 37.01 Å, lDDT 0.83, secondary-structure agreement Q3 79%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.43. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6TAZ_B at mean pLDDT 74.4, but the deposited structure scores TM-score 0.48 against that prediction — 73% of residues confidently wrong. A disorder / coiled-coil ambiguity.
Protein timeless homolog is a 141-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 74.4 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 10.85 Å, lDDT 0.72, secondary-structure agreement Q3 92%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.87. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6ZE9_A at mean pLDDT 81.2, but the deposited structure scores TM-score 0.36 against that prediction — 80% of residues confidently wrong. A wrong biological context.
Vam6/Vps39-like protein is a 44-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 81.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 13.64 Å, lDDT 0.39, secondary-structure agreement Q3 54%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.24. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6WK1_Y at mean pLDDT 96.4, but the deposited structure scores TM-score 0.18 against that prediction — 79% of residues confidently wrong. A wrong biological context.
Actin, cytoplasmic 2 is a 23-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.4 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.18, Cα-RMSD 8.56 Å, lDDT 0.64, secondary-structure agreement Q3 47%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.08. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6LNI_A at mean pLDDT 85.2, but the deposited structure scores TM-score 0.29 against that prediction — 93% of residues confidently wrong. A native fold → amyloid.
Major prion protein is a 210-residue chain; the deposition resolves 10 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 12.84 Å, lDDT 0.40, secondary-structure agreement Q3 10%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.11. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6W6W_C at mean pLDDT 92.2, but the deposited structure scores TM-score 0.48 against that prediction — 98% of residues confidently wrong. A right fold, wrong conformation/assembly.
CST complex subunit STN1 is a 374-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 92.2 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 20.42 Å, lDDT 0.86, secondary-structure agreement Q3 79%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.31. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6ULK_C at mean pLDDT 96.7, but the deposited structure scores TM-score 0.29 against that prediction — 50% of residues confidently wrong. A wrong biological context.
GLY-ALA-ASP-GLY-VAL-GLY-LYS-SER-ALA-LEU is a 10-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.7 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 4.76 Å, lDDT 0.62, secondary-structure agreement Q3 50%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.36. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2020-05-21MR78 antobody heavy chain✓ no confident missTM 0.92pLDDT 97FRAUD 0.08
1 structure analysed this release. AlphaFold's largest disagreement was MR78 antobody heavy chain (TM 0.92) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6WG3_C at mean pLDDT 85.7, but the deposited structure scores TM-score 0.32 against that prediction — 90% of residues confidently wrong. A right fold, wrong conformation/assembly.
Double-strand-break repair protein rad21 homolog is a 631-residue chain; the deposition resolves 7 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.7 signals high confidence, yet the accuracy metrics disagree: TM-score 0.32, Cα-RMSD 45.23 Å, lDDT 0.83, secondary-structure agreement Q3 92%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.58. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2020-05-13Receptor-type tyrosine-protein phosphatase F✓ no confident missTM 0.51pLDDT 85FRAUD 0.83
130 structures analysed this release. AlphaFold's largest disagreement was Receptor-type tyrosine-protein phosphatase F (TM 0.51) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6TR0_A at mean pLDDT 86.5, but the deposited structure scores TM-score 0.46 against that prediction — 90% of residues confidently wrong. A right fold, wrong conformation/assembly.
Splicing factor U2AF 65 kDa subunit is a 205-residue chain annotated as RRM (RNA recognition motif) domain (3). This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.46, Cα-RMSD 15.77 Å, lDDT 0.77, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.71. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6Y95_A at mean pLDDT 85.5, but the deposited structure scores TM-score 0.39 against that prediction — 85% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin is a 148-residue chain annotated as EF-hand (1). This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.39, Cα-RMSD 13.18 Å, lDDT 0.64, secondary-structure agreement Q3 90%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.26. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2020-04-22NF-kappa-B-repressing factor✓ no confident missTM 0.58pLDDT 67FRAUD 0.25
139 structures analysed this release. AlphaFold's largest disagreement was NF-kappa-B-repressing factor (TM 0.58) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6L54_A at mean pLDDT 83.1, but the deposited structure scores TM-score 0.50 against that prediction — 49% of residues confidently wrong. A wrong biological context.
Serine/threonine-protein kinase SMG1 is a 3661-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 17.83 Å, lDDT 0.53, secondary-structure agreement Q3 69%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.36. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 71% novel to AlphaFold's training cutoff.
AlphaFold modelled 6V5C_B at mean pLDDT 87.3, but the deposited structure scores TM-score 0.37 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
Microprocessor complex subunit DGCR8 is a 532-residue chain annotated as dsRNA-binding domain-like; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.37, Cα-RMSD 18.54 Å, lDDT 0.60, secondary-structure agreement Q3 73%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.46. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2020-04-01Engulfment and cell motility protein 1✓ no confident missTM 0.67pLDDT 90FRAUD 0.65
105 structures analysed this release. AlphaFold's largest disagreement was Engulfment and cell motility protein 1 (TM 0.67) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6UUN_P at mean pLDDT 70.8, but the deposited structure scores TM-score 0.49 against that prediction — 5% of residues confidently wrong. A wrong biological context.
ADM is a 53-residue chain; the deposition resolves 7 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 70.8 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 4.76 Å, lDDT 0.79, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.46. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
87 structures analysed this release. AlphaFold's largest disagreement was 26S proteasome non-ATPase regulatory subunit 4 (TM 0.50) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2020-03-11DNA excision repair protein ERCC-1✓ no confident missTM 0.62pLDDT 92FRAUD 0.65
97 structures analysed this release. AlphaFold's largest disagreement was DNA excision repair protein ERCC-1 (TM 0.62) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6JK8_A at mean pLDDT 87.0, but the deposited structure scores TM-score 0.49 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
Insulin-like growth factor 1 receptor is a 1367-residue chain annotated as Growth factor receptor domain-like; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 16.56 Å, lDDT 0.76, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.49. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6UCK_A at mean pLDDT 76.6, but the deposited structure scores TM-score 0.43 against that prediction — 31% of residues confidently wrong. A native fold → amyloid.
Islet amyloid polypeptide is a 72-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 5.38 Å, lDDT 0.83, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.49. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6U39_A at mean pLDDT 87.6, but the deposited structure scores TM-score 0.50 against that prediction — 97% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 148-residue chain annotated as EF-hand; the deposition resolves 20 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 19.54 Å, lDDT 0.74, secondary-structure agreement Q3 91%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.01. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6LTH_N at mean pLDDT 89.0, but the deposited structure scores TM-score 0.38 against that prediction — 96% of residues confidently wrong. A right fold, wrong conformation/assembly.
SWI/SNF complex subunit SMARCC2 is a 1214-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 28.17 Å, lDDT 0.82, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.82. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2020-02-07Subunit B✓ no confident missTM 0.69pLDDT 93FRAUD 0.23
4 structures analysed this release. AlphaFold's largest disagreement was Subunit B (TM 0.69) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2020-02-0510-1074 Fab Light ChainRight fold, wrong conformation/assemblyTM 0.48pLDDT 94FRAUD 0.54
AlphaFold modelled 6V8Z_E at mean pLDDT 94.1, but the deposited structure scores TM-score 0.48 against that prediction — 87% of residues confidently wrong. A right fold, wrong conformation/assembly.
10-1074 Fab Light Chain is a 211-residue chain annotated as Immunoglobulins (2); the deposition resolves 18 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 94.1 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 9.76 Å, lDDT 0.73, secondary-structure agreement Q3 71%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.54. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6JAT_B at mean pLDDT 96.6, but the deposited structure scores TM-score 0.23 against that prediction — 75% of residues confidently wrong. A wrong biological context.
Actin, gamma-enteric smooth muscle is a 22-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.23, Cα-RMSD 7.68 Å, lDDT 0.64, secondary-structure agreement Q3 56%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.07. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6V63_Y at mean pLDDT 96.4, but the deposited structure scores TM-score 0.22 against that prediction — 79% of residues confidently wrong. A right fold, wrong conformation/assembly.
Actin, cytoplasmic 1 is a 23-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.4 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.22, Cα-RMSD 8.55 Å, lDDT 0.64, secondary-structure agreement Q3 68%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.19. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6KN7_T at mean pLDDT 93.9, but the deposited structure scores TM-score 0.49 against that prediction — 97% of residues confidently wrong. A right fold, wrong conformation/assembly.
Troponin T, cardiac muscle is a 186-residue chain annotated as Troponin coil-coiled subunits; the deposition resolves 29 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 93.9 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 27.93 Å, lDDT 0.92, secondary-structure agreement Q3 91%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.52. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2020-01-10Subunit Alpha_46-219✓ no confident missTM 0.50pLDDT 74FRAUD 0.48
7 structures analysed this release. AlphaFold's largest disagreement was Subunit Alpha_46-219 (TM 0.50) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2020-01-08Guanine nucleotide-binding protein G(s) subunit alpha isoforms short✓ no confident missTM 0.86pLDDT 93FRAUD 0.14
32 structures analysed this release. AlphaFold's largest disagreement was Guanine nucleotide-binding protein G(s) subunit alpha isoforms short (TM 0.86) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2020-01-01Protein ECT2✓ no confident missTM 0.61pLDDT 87FRAUD 0.76
76 structures analysed this release. AlphaFold's largest disagreement was Protein ECT2 (TM 0.61) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6MAK_B at mean pLDDT 71.8, but the deposited structure scores TM-score 0.48 against that prediction — 0% of residues confidently wrong. A wrong biological context.
BRD1 protein is a 52-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 71.8 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 2.77 Å, lDDT 0.88, secondary-structure agreement Q3 41%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.41. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2019-12-18DET1- and DDB1-associated protein 1Wrong biological contextTM 0.26pLDDT 72FRAUD 0.59
AlphaFold modelled 6UD7_D at mean pLDDT 72.1, but the deposited structure scores TM-score 0.26 against that prediction — 67% of residues confidently wrong. A wrong biological context.
DET1- and DDB1-associated protein 1 is a 101-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.1 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.26, Cα-RMSD 17.72 Å, lDDT 0.73, secondary-structure agreement Q3 82%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.55. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6SYT_A at mean pLDDT 84.0, but the deposited structure scores TM-score 0.50 against that prediction — 43% of residues confidently wrong. A right fold, wrong conformation/assembly.
SMG1,Serine/threonine-protein kinase SMG1,SMG1,Serine/threonine-protein kinase SMG1,SMG1,Serine/threonine-protein kinase SMG1,SMG1,Serine/threonine-protein kinase SMG1,SMG1,Serine/threonine-protein kinase SMG1 is a 3712-residue chain; the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 84.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.50, Cα-RMSD 14.76 Å, lDDT 0.78, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.50. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 74% novel to AlphaFold's training cutoff.
AlphaFold modelled 6V01_B at mean pLDDT 86.3, but the deposited structure scores TM-score 0.49 against that prediction — 83% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-1 is a 149-residue chain; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 11.72 Å, lDDT 0.68, secondary-structure agreement Q3 81%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.38. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6TYZ_B at mean pLDDT 77.2, but the deposited structure scores TM-score 0.35 against that prediction — 0% of residues confidently wrong. A wrong biological context.
GLU-ARG-LYS-ARG-ILE-LEU-PRO-THR-TRP-MET-LEU-ALA is a 16-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 2.10 Å, lDDT 0.94, secondary-structure agreement Q3 67%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.54. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6NL3_A at mean pLDDT 87.8, but the deposited structure scores TM-score 0.27 against that prediction — 84% of residues confidently wrong. A disorder / coiled-coil ambiguity.
Cytochrome c oxidase assembly factor 6 homolog is a 81-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.27, Cα-RMSD 13.74 Å, lDDT 0.48, secondary-structure agreement Q3 57%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.34. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is 100% novel to AlphaFold's training cutoff.
2019-11-13DET1- and DDB1-associated protein 1Wrong biological contextTM 0.29pLDDT 75FRAUD 0.63
AlphaFold modelled 6Q0W_E at mean pLDDT 74.8, but the deposited structure scores TM-score 0.29 against that prediction — 81% of residues confidently wrong. A wrong biological context.
DET1- and DDB1-associated protein 1 is a 126-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 74.8 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 17.67 Å, lDDT 0.74, secondary-structure agreement Q3 69%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.51. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2019-11-06Folliculin✓ no confident missTM 0.51pLDDT 91FRAUD 0.53
56 structures analysed this release. AlphaFold's largest disagreement was Folliculin (TM 0.51) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6HVK_A at mean pLDDT 85.4, but the deposited structure scores TM-score 0.38 against that prediction — 8% of residues confidently wrong. A whole-fold mismatch.
Urotensin-2 receptor is a 14-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 2.71 Å, lDDT 0.69, secondary-structure agreement Q3 25%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.21. Mechanistically, secondary structure and tertiary packing both diverge from the prediction — AlphaFold committed to the wrong fold. The sequence is within AlphaFold's training distribution.
2019-10-23Paxillin✓ no confident missTM 0.57pLDDT 92FRAUD 0.28
72 structures analysed this release. AlphaFold's largest disagreement was Paxillin (TM 0.57) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2019-10-17tau protein✓ no confident missTM 0.10pLDDT 54FRAUD 0.54
1 structure analysed this release. AlphaFold's largest disagreement was tau protein (TM 0.10) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6SB0_O at mean pLDDT 75.0, but the deposited structure scores TM-score 0.49 against that prediction — 66% of residues confidently wrong. A wrong biological context.
Proline-rich AKT1 substrate 1 is a 256-residue chain; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 75.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 17.00 Å, lDDT 0.82, secondary-structure agreement Q3 97%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.61. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6MN6_A at mean pLDDT 85.3, but the deposited structure scores TM-score 0.47 against that prediction — 92% of residues confidently wrong. A right fold, wrong conformation/assembly.
Metal transporter CNNM3 is a 365-residue chain annotated as CBS-domain (3); the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 18.19 Å, lDDT 0.88, secondary-structure agreement Q3 90%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.63. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2019-10-02Aquaporin-1✓ no confident missTM 0.68pLDDT 91FRAUD 0.31
79 structures analysed this release. AlphaFold's largest disagreement was Aquaporin-1 (TM 0.68) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2019-09-25Alpha-synuclein✓ no confident missTM 0.28pLDDT 85FRAUD 0.81
84 structures analysed this release. AlphaFold's largest disagreement was Alpha-synuclein (TM 0.28) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2019-09-18DET1- and DDB1-associated protein 1Wrong biological contextTM 0.30pLDDT 74FRAUD 0.58
AlphaFold modelled 6PAI_E at mean pLDDT 74.0, but the deposited structure scores TM-score 0.30 against that prediction — 72% of residues confidently wrong. A wrong biological context.
DET1- and DDB1-associated protein 1 is a 102-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 74.0 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.30, Cα-RMSD 15.65 Å, lDDT 0.75, secondary-structure agreement Q3 86%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.63. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6Q6G_M at mean pLDDT 76.8, but the deposited structure scores TM-score 0.25 against that prediction — 64% of residues confidently wrong. A wrong biological context.
Anaphase-promoting complex subunit 13 is a 74-residue chain annotated as Anaphase-promoting complex subunit 13; the deposition resolves 21 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.25, Cα-RMSD 20.75 Å, lDDT 0.73, secondary-structure agreement Q3 80%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.71. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6PXV_A at mean pLDDT 87.5, but the deposited structure scores TM-score 0.49 against that prediction — 91% of residues confidently wrong. A right fold, wrong conformation/assembly.
Insulin receptor is a 1354-residue chain annotated as Receptor L-domain (3); the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 19.59 Å, lDDT 0.90, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.64. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2019-08-28Transient receptor potential cation channel subfamily A member 1Wrong biological contextTM 0.35pLDDT 92FRAUD 0.34
AlphaFold modelled 6HC8_E at mean pLDDT 91.6, but the deposited structure scores TM-score 0.35 against that prediction — 75% of residues confidently wrong. A wrong biological context.
Transient receptor potential cation channel subfamily A member 1 is a 27-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.6 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 5.94 Å, lDDT 0.69, secondary-structure agreement Q3 50%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.61. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6RUR_B at mean pLDDT 79.2, but the deposited structure scores TM-score 0.40 against that prediction — 77% of residues confidently wrong. A wrong biological context.
Complement C3 is a 915-residue chain; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.40, Cα-RMSD 28.02 Å, lDDT 0.78, secondary-structure agreement Q3 86%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.64. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6KBO_B at mean pLDDT 87.3, but the deposited structure scores TM-score 0.33 against that prediction — 100% of residues confidently wrong. A wrong biological context.
Heparin cofactor 2 is a 28-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.33, Cα-RMSD 8.93 Å, lDDT 0.42, secondary-structure agreement Q3 36%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.27. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2019-08-07Retinol-binding protein 2✓ no confident missTM 0.58pLDDT 97FRAUD 0.72
83 structures analysed this release. AlphaFold's largest disagreement was Retinol-binding protein 2 (TM 0.58) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2019-07-31Protocadherin-15✓ no confident missTM 0.57pLDDT 83FRAUD 0.61
82 structures analysed this release. AlphaFold's largest disagreement was Protocadherin-15 (TM 0.57) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6S5T_B at mean pLDDT 86.9, but the deposited structure scores TM-score 0.38 against that prediction — 86% of residues confidently wrong. A right fold, wrong conformation/assembly.
Calmodulin-2 is a 149-residue chain annotated as EF-hand; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 86.9 signals high confidence, yet the accuracy metrics disagree: TM-score 0.38, Cα-RMSD 10.63 Å, lDDT 0.66, secondary-structure agreement Q3 87%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.32. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6A8Y_A at mean pLDDT 92.9, but the deposited structure scores TM-score 0.34 against that prediction — 96% of residues confidently wrong. A disorder / coiled-coil ambiguity.
YR26_SDS is a 26-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 92.9 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.34, Cα-RMSD 7.38 Å, lDDT 0.44, secondary-structure agreement Q3 38%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.09. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is 100% novel to AlphaFold's training cutoff.
2019-07-10X-box-binding protein 1Wrong biological contextTM 0.18pLDDT 71FRAUD 0.14
AlphaFold modelled 6R7Q_1 at mean pLDDT 71.3, but the deposited structure scores TM-score 0.18 against that prediction — 4% of residues confidently wrong. A wrong biological context.
X-box-binding protein 1 is a 24-residue chain; the deposition resolves 88 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 71.3 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.18, Cα-RMSD 3.25 Å, lDDT 0.81, secondary-structure agreement Q3 100%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.33. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6Q2X_L at mean pLDDT 72.9, but the deposited structure scores TM-score 0.43 against that prediction — 8% of residues confidently wrong. A wrong biological context.
Transcriptional coactivator YAP1 is a 41-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.9 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.43, Cα-RMSD 4.14 Å, lDDT 0.79, secondary-structure agreement Q3 85%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.73. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2019-06-26Heat shock protein beta-1Wrong biological contextTM 0.39pLDDT 70FRAUD 0.65
AlphaFold modelled 6DV5_A at mean pLDDT 70.1, but the deposited structure scores TM-score 0.39 against that prediction — 50% of residues confidently wrong. A wrong biological context.
Heat shock protein beta-1 is a 205-residue chain annotated as Hsp20 chaperone-like; the deposition resolves 24 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 70.1 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.39, Cα-RMSD 26.91 Å, lDDT 0.41, secondary-structure agreement Q3 67%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.79. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2019-06-19Double-strand-break repair protein rad21 homologWrong biological contextTM 0.36pLDDT 92FRAUD 0.07
AlphaFold modelled 6RRC_B at mean pLDDT 92.5, but the deposited structure scores TM-score 0.36 against that prediction — 0% of residues confidently wrong. A wrong biological context.
Double-strand-break repair protein rad21 homolog is a 25-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 92.5 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 1.41 Å, lDDT 0.88, secondary-structure agreement Q3 44%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.51. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6DRD_M at mean pLDDT 78.1, but the deposited structure scores TM-score 0.36 against that prediction — 74% of residues confidently wrong. A wrong biological context.
DNA-directed RNA polymerase II subunit GRINL1A is a 62-residue chain; the deposition resolves 13 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 27.96 Å, lDDT 0.57, secondary-structure agreement Q3 69%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.33. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6NZN_A at mean pLDDT 80.2, but the deposited structure scores TM-score 0.28 against that prediction — 66% of residues confidently wrong. A native fold → amyloid.
Glucagon is a 29-residue chain; the deposition resolves 16 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 80.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.28, Cα-RMSD 17.20 Å, lDDT 0.45, secondary-structure agreement Q3 7%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.83. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6O9L_3 at mean pLDDT 85.3, but the deposited structure scores TM-score 0.41 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
CDK-activating kinase assembly factor MAT1 is a 309-residue chain; the deposition resolves 33 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 85.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.41, Cα-RMSD 31.77 Å, lDDT 0.60, secondary-structure agreement Q3 77%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.35. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6Q8I_C at mean pLDDT 83.8, but the deposited structure scores TM-score 0.48 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
Protein Red is a 557-residue chain; the deposition resolves 16 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 83.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 8.37 Å, lDDT 0.76, secondary-structure agreement Q3 68%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.19. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6GR8_B at mean pLDDT 79.5, but the deposited structure scores TM-score 0.48 against that prediction — 52% of residues confidently wrong. A wrong biological context.
Inner centromere protein is a 70-residue chain annotated as INCENP ARK binding region-like; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 7.53 Å, lDDT 0.83, secondary-structure agreement Q3 100%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.85. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2019-05-08Calmodulin-1✓ no confident missTM 0.58pLDDT 86FRAUD 0.31
101 structures analysed this release. AlphaFold's largest disagreement was Calmodulin-1 (TM 0.58) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6H8C_B at mean pLDDT 79.8, but the deposited structure scores TM-score 0.23 against that prediction — 63% of residues confidently wrong. A wrong biological context.
Ubiquitin-like modifier-activating enzyme 5 is a 19-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 79.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.23, Cα-RMSD 6.91 Å, lDDT 0.69, secondary-structure agreement Q3 68%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.31. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2019-04-29UbcH5c✓ no confident missTM 0.95pLDDT 95FRAUD 0.05
7 structures analysed this release. AlphaFold's largest disagreement was UbcH5c (TM 0.95) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
125 structures analysed this release. AlphaFold's largest disagreement was Nuclear RNA export factor 1 (TM 0.63) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6QW6_X at mean pLDDT 74.1, but the deposited structure scores TM-score 0.36 against that prediction — 67% of residues confidently wrong. A wrong biological context.
U4/U6.U5 small nuclear ribonucleoprotein 27 kDa protein is a 155-residue chain; the deposition resolves 39 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 74.1 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 7.45 Å, lDDT 0.72, secondary-structure agreement Q3 55%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.28. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
AlphaFold modelled 6DKP_C at mean pLDDT 89.3, but the deposited structure scores TM-score 0.29 against that prediction — 40% of residues confidently wrong. A wrong biological context.
Melanoma antigen recognized by T-cells 1 is a 10-residue chain; the deposition resolves 5 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.29, Cα-RMSD 4.20 Å, lDDT 0.44, secondary-structure agreement Q3 0%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.10. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6EDS_C at mean pLDDT 72.3, but the deposited structure scores TM-score 0.27 against that prediction — 42% of residues confidently wrong. A wrong biological context.
Glucagon is a 29-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 72.3 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.27, Cα-RMSD 7.33 Å, lDDT 0.47, secondary-structure agreement Q3 50%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.09. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2019-03-27Microtubule-associated protein tau✓ no confident missTM 0.25pLDDT 68FRAUD 0.65
115 structures analysed this release. AlphaFold's largest disagreement was Microtubule-associated protein tau (TM 0.25) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2019-03-20PRELI domain containing protein 3B✓ no confident missTM 0.69pLDDT 84FRAUD 0.65
60 structures analysed this release. AlphaFold's largest disagreement was PRELI domain containing protein 3B (TM 0.69) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 5ZT0_G at mean pLDDT 82.2, but the deposited structure scores TM-score 0.20 against that prediction — 68% of residues confidently wrong. A right fold, wrong conformation/assembly.
Protein phosphatase 1 regulatory subunit 3B is a 75-residue chain; the deposition resolves 10 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 82.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.20, Cα-RMSD 5.04 Å, lDDT 0.75, secondary-structure agreement Q3 79%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.56. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is 100% novel to AlphaFold's training cutoff.
2019-03-06Coagulation factor XI✓ no confident missTM 0.68pLDDT 89FRAUD 0.53
27 structures analysed this release. AlphaFold's largest disagreement was Coagulation factor XI (TM 0.68) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6ICT_E at mean pLDDT 96.3, but the deposited structure scores TM-score 0.15 against that prediction — 79% of residues confidently wrong. A right fold, wrong conformation/assembly.
Actin, cytoplasmic 1 is a 23-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 96.3 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.15, Cα-RMSD 8.56 Å, lDDT 0.67, secondary-structure agreement Q3 63%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.16. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6QDV_t at mean pLDDT 89.0, but the deposited structure scores TM-score 0.44 against that prediction — 96% of residues confidently wrong. A right fold, wrong conformation/assembly.
Pre-mRNA-processing factor 19 is a 504-residue chain; the deposition resolves 54 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.0 signals high confidence, yet the accuracy metrics disagree: TM-score 0.44, Cα-RMSD 19.61 Å, lDDT 0.91, secondary-structure agreement Q3 94%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.86. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2019-02-13Excision repair cross-complementing rodent repair deficiency, complementation group 6 variant✓ no confident missTM 0.72pLDDT 59FRAUD 0.16
46 structures analysed this release. AlphaFold's largest disagreement was Excision repair cross-complementing rodent repair deficiency, complementation group 6 variant (TM 0.72) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2019-02-06Protein RD3✓ no confident missTM 0.61pLDDT 90FRAUD 0.62
56 structures analysed this release. AlphaFold's largest disagreement was Protein RD3 (TM 0.61) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2019-01-30Guanine nucleotide-binding protein G(i) subunit alpha-1✓ no confident missTM 0.81pLDDT 93FRAUD 0.18
45 structures analysed this release. AlphaFold's largest disagreement was Guanine nucleotide-binding protein G(i) subunit alpha-1 (TM 0.81) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6HVO_D at mean pLDDT 80.8, but the deposited structure scores TM-score 0.36 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
DNA polymerase delta subunit 4 is a 19-residue chain; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 80.8 signals high confidence, yet the accuracy metrics disagree: TM-score 0.36, Cα-RMSD 2.00 Å, lDDT 0.81, secondary-structure agreement Q3 83%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.15. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6J12_A at mean pLDDT 87.3, but the deposited structure scores TM-score 0.35 against that prediction — 79% of residues confidently wrong. A disorder / coiled-coil ambiguity.
Heparin cofactor 2 is a 28-residue chain. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 87.3 signals high confidence, yet the accuracy metrics disagree: TM-score 0.35, Cα-RMSD 9.04 Å, lDDT 0.46, secondary-structure agreement Q3 36%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.11. Mechanistically, the sequence lacks a single well-defined fold, so AlphaFold's one static answer cannot match the deposited state. The sequence is within AlphaFold's training distribution.
2019-01-09Nuclear receptor coactivator 1✓ no confident missTM 0.79pLDDT 47FRAUD 0.10
32 structures analysed this release. AlphaFold's largest disagreement was Nuclear receptor coactivator 1 (TM 0.79) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6DK5_A at mean pLDDT 77.1, but the deposited structure scores TM-score 0.28 against that prediction — 5% of residues confidently wrong. A wrong biological context.
Endothelin-1 is a 21-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 77.1 signals high confidence, yet the accuracy metrics disagree: TM-score 0.28, Cα-RMSD 4.12 Å, lDDT 0.74, secondary-structure agreement Q3 81%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.90. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6DXR_A at mean pLDDT 76.6, but the deposited structure scores TM-score 0.48 against that prediction — 71% of residues confidently wrong. A wrong biological context.
Apolipoprotein C-I is a 83-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 76.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 10.47 Å, lDDT 0.89, secondary-structure agreement Q3 86%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.71. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6MBK_Y at mean pLDDT 95.5, but the deposited structure scores TM-score 0.24 against that prediction — 67% of residues confidently wrong. A right fold, wrong conformation/assembly.
Actin peptide is a 15-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 95.5 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.24, Cα-RMSD 6.73 Å, lDDT 0.66, secondary-structure agreement Q3 60%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.15. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2018-12-12DET1- and DDB1-associated protein 1Wrong biological contextTM 0.22pLDDT 73FRAUD 0.23
AlphaFold modelled 6DSZ_C at mean pLDDT 73.0, but the deposited structure scores TM-score 0.22 against that prediction — 58% of residues confidently wrong. A wrong biological context.
DET1- and DDB1-associated protein 1 is a 19-residue chain; the deposition resolves 4 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 73.0 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.22, Cα-RMSD 5.25 Å, lDDT 0.76, secondary-structure agreement Q3 100%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.00. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is within AlphaFold's training distribution.
2018-12-05Calmodulin-1✓ no confident missTM 0.52pLDDT 88FRAUD 0.57
127 structures analysed this release. AlphaFold's largest disagreement was Calmodulin-1 (TM 0.52) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6MZM_K at mean pLDDT 88.2, but the deposited structure scores TM-score 0.47 against that prediction — 98% of residues confidently wrong. A right fold, wrong conformation/assembly.
Transcription initiation factor TFIID subunit 8, TAF8 is a 310-residue chain annotated as Core histone-like; the deposition resolves 17 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 88.2 signals high confidence, yet the accuracy metrics disagree: TM-score 0.47, Cα-RMSD 27.04 Å, lDDT 0.75, secondary-structure agreement Q3 81%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.63. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2018-11-21Fab 2g12, heavy chain✓ no confident missTM 0.55pLDDT 92FRAUD 0.69
193 structures analysed this release. AlphaFold's largest disagreement was Fab 2g12, heavy chain (TM 0.55) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6GK3_A at mean pLDDT 97.9, but the deposited structure scores TM-score 0.22 against that prediction — 100% of residues confidently wrong. A native fold → amyloid.
Beta-2-microglobulin is a 64-residue chain; the deposition resolves 8 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 97.9 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.22, Cα-RMSD 22.00 Å, lDDT 0.54, secondary-structure agreement Q3 50%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is +0.22. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
2018-11-07Transforming growth factor beta-1✓ no confident missTM 0.62pLDDT 87FRAUD 0.76
47 structures analysed this release. AlphaFold's largest disagreement was Transforming growth factor beta-1 (TM 0.62) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-10-31Fab 2G12, heavy chain✓ no confident missTM 0.54pLDDT 92FRAUD 0.68
28 structures analysed this release. AlphaFold's largest disagreement was Fab 2G12, heavy chain (TM 0.54) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-10-26E6AP HECT Domain✓ no confident missTM 0.92pLDDT 81FRAUD 0.06
3 structures analysed this release. AlphaFold's largest disagreement was E6AP HECT Domain (TM 0.92) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6A44_A at mean pLDDT 91.4, but the deposited structure scores TM-score 0.31 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
RNA-directed DNA polymerase homolog (R1),Polyubiquitin-C is a 302-residue chain annotated as Endonuclease/exonuclease/phosphatase (3). This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 91.4 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.31, Cα-RMSD 0.49 Å, lDDT 0.96, secondary-structure agreement Q3 93%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.34. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2018-10-17Calmodulin-1✓ no confident missTM 0.50pLDDT 86FRAUD 0.62
48 structures analysed this release. AlphaFold's largest disagreement was Calmodulin-1 (TM 0.50) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-10-10Microtubule-associated protein tau✓ no confident missTM 0.27pLDDT 68FRAUD 0.66
147 structures analysed this release. AlphaFold's largest disagreement was Microtubule-associated protein tau (TM 0.27) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-10-03Heat shock protein HSP 90-alpha✓ no confident missTM 0.73pLDDT 85FRAUD 0.23
59 structures analysed this release. AlphaFold's largest disagreement was Heat shock protein HSP 90-alpha (TM 0.73) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6E28_C at mean pLDDT 89.5, but the deposited structure scores TM-score 0.48 against that prediction — 93% of residues confidently wrong. A right fold, wrong conformation/assembly.
Caspase recruitment domain-containing protein 9 is a 97-residue chain; the deposition resolves 2 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.5 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 19.56 Å, lDDT 0.89, secondary-structure agreement Q3 84%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.62. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2018-09-19Histone H3✓ no confident missTM 0.18pLDDT 62FRAUD 0.15
33 structures analysed this release. AlphaFold's largest disagreement was Histone H3 (TM 0.18) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-09-12Microtubule-associated protein tau✓ no confident missTM 0.25pLDDT 67FRAUD 0.62
13 structures analysed this release. AlphaFold's largest disagreement was Microtubule-associated protein tau (TM 0.25) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-09-05Transcription elongation factor SPT5✓ no confident missTM 0.58pLDDT 90FRAUD 0.57
28 structures analysed this release. AlphaFold's largest disagreement was Transcription elongation factor SPT5 (TM 0.58) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-08-29FYN✓ no confident missTM 0.80pLDDT 81FRAUD 0.16
33 structures analysed this release. AlphaFold's largest disagreement was FYN (TM 0.80) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6GMH_Z at mean pLDDT 89.6, but the deposited structure scores TM-score 0.49 against that prediction — 88% of residues confidently wrong. A right fold, wrong conformation/assembly.
Transcription elongation factor SPT5 is a 1087-residue chain; the deposition resolves 23 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 89.6 signals high confidence, yet the accuracy metrics disagree: TM-score 0.49, Cα-RMSD 14.94 Å, lDDT 0.86, secondary-structure agreement Q3 83%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.47. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
AlphaFold modelled 6H25_K at mean pLDDT 78.4, but the deposited structure scores TM-score 0.48 against that prediction — 32% of residues confidently wrong. A wrong biological context.
M-phase phosphoprotein 6 is a 166-residue chain; the deposition resolves 12 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 78.4 signals high confidence, yet the accuracy metrics disagree: TM-score 0.48, Cα-RMSD 3.80 Å, lDDT 0.65, secondary-structure agreement Q3 81%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.51. Mechanistically, the experiment captured a complex or assembly state that a single-chain monomer prediction cannot anticipate. The sequence is 100% novel to AlphaFold's training cutoff.
2018-08-08Alpha-synuclein✓ no confident missTM 0.26pLDDT 86FRAUD 0.80
50 structures analysed this release. AlphaFold's largest disagreement was Alpha-synuclein (TM 0.26) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-08-01Dynamin-1✓ no confident missTM 0.68pLDDT 87FRAUD 0.48
12 structures analysed this release. AlphaFold's largest disagreement was Dynamin-1 (TM 0.68) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-07-25Alpha N-terminal protein methyltransferase 1B✓ no confident missTM 0.96pLDDT 90FRAUD 0.04
14 structures analysed this release. AlphaFold's largest disagreement was Alpha N-terminal protein methyltransferase 1B (TM 0.96) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-07-18Inositol 1,4,5-trisphosphate receptor type 3✓ no confident missTM 0.76pLDDT 73FRAUD 0.18
23 structures analysed this release. AlphaFold's largest disagreement was Inositol 1,4,5-trisphosphate receptor type 3 (TM 0.76) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-07-11Alpha-synuclein✓ no confident missTM 0.29pLDDT 84FRAUD 0.80
22 structures analysed this release. AlphaFold's largest disagreement was Alpha-synuclein (TM 0.29) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-07-04Protein S100-A8✓ no confident missTM 0.93pLDDT 93FRAUD 0.07
19 structures analysed this release. AlphaFold's largest disagreement was Protein S100-A8 (TM 0.93) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-06-27Immunoglobulin G PGT151 Fab, Heavy chain✓ no confident missTM 0.81pLDDT 87FRAUD 0.17
12 structures analysed this release. AlphaFold's largest disagreement was Immunoglobulin G PGT151 Fab, Heavy chain (TM 0.81) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-06-20Peptide from Microtubule-associated protein tau✓ no confident missTM 0.31pLDDT 47FRAUD 0.21
10 structures analysed this release. AlphaFold's largest disagreement was Peptide from Microtubule-associated protein tau (TM 0.31) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold modelled 6GK7_A at mean pLDDT 70.5, but the deposited structure scores TM-score 0.31 against that prediction — 0% of residues confidently wrong. A native fold → amyloid.
HUMAN TAU PEPTIDE A8119 RESIDUES 299-318 is a 11-residue chain annotated as Immunoglobulins (2); the deposition resolves 3 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 70.5 signals moderate confidence, yet the accuracy metrics disagree: TM-score 0.31, Cα-RMSD 2.45 Å, lDDT 0.83, secondary-structure agreement Q3 36%.
Confidence is mis-calibrated here — the pLDDT↔lDDT correlation is -0.28. Mechanistically, the deposited coordinates are a self-templating cross-β amyloid assembly — a non-native polymer state AlphaFold's single-chain model has no way to represent. The sequence is within AlphaFold's training distribution.
2018-06-0635O22 heavy chainRight fold, wrong conformation/assemblyTM 0.45pLDDT 93FRAUD 0.04
AlphaFold modelled 6DE7_D at mean pLDDT 93.0, but the deposited structure scores TM-score 0.45 against that prediction — 0% of residues confidently wrong. A right fold, wrong conformation/assembly.
35O22 heavy chain is a 243-residue chain annotated as gp120 core; the deposition resolves 6 chains. This is the fold AlphaFold predicts, and it does so with high confidence.
In the viewer, residues are coloured by their Cα deviation from the AlphaFold model — red marks where the deposited coordinates moved furthest from the prediction; toggle the AlphaFold “ghost” to overlay what it predicted.
AlphaFold's mean pLDDT of 93.0 signals near-maximal confidence, yet the accuracy metrics disagree: TM-score 0.45, Cα-RMSD 0.82 Å, lDDT 0.90, secondary-structure agreement Q3 70%.
Confidence is only weakly informative — the pLDDT↔lDDT correlation is +0.61. Mechanistically, the local fold is largely correct but the global arrangement is not — the hallmark of a conformational-state or inter-domain-geometry error rather than a misfold. The sequence is within AlphaFold's training distribution.
2018-05-30Beclin-1✓ no confident missTM 0.78pLDDT 77FRAUD 0.17
4 structures analysed this release. AlphaFold's largest disagreement was Beclin-1 (TM 0.78) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-05-23Beclin-1✓ no confident missTM 0.76pLDDT 77FRAUD 0.19
11 structures analysed this release. AlphaFold's largest disagreement was Beclin-1 (TM 0.76) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
2018-05-16Dihydrofolate reductase✓ no confident missTM 0.98pLDDT 96FRAUD 0.02
1 structure analysed this release. AlphaFold's largest disagreement was Dihydrofolate reductase (TM 0.98) — but nothing was confidently wrong, so there is no overconfident failure to report this week.
AlphaFold is genuinely good at compact, single-chain globular folds — which is exactly why these four are so telling. Each is a regular folded protein, not an amyloid, a membrane protein, or a disordered chain. AlphaFold gets the fold essentially right, then commits — often at very high confidence — to the wrong conformation, curvature, or inter-domain pose. The failure is subtle, and its rarity is itself the finding.
PP2A scaffold subunit A (PR65) PPP2R1A
Right fold, wrong curvature (flexible α-solenoid)globularflexible scaffoldtumour suppressorP30153 ↗
AlphaFold assigns mean pLDDT 96.8 — near-maximal confidence — to a 15-repeat HEAT solenoid whose global curvature is wrong for 99% of its residues: a textbook case of high confidence in the wrong spring state.
Worst deposition: 8UWB_C ·
drag to rotate · scroll to zoom · toggle the AlphaFold “ghost” to see what it predicted.
0.29
worst TM-score
96.76
mean pLDDT
0.77
FRAUD score
×26
structures
1/26
conf. wrong
0%
novelty
🧬 Structure & architecture
The PP2A A/scaffold subunit (PR65) is built entirely from 15 tandem HEAT repeats, each an antiparallel pair of ~39-residue α-helices (a short 'A' helix and a longer 'B' helix) joined by a turn. Successive repeats stack side-by-side into a superhelical, horseshoe-shaped α-solenoid ~90 Å long, with no β-structure or globular core — architecturally distinct from most folds in the AlphaFraud set.
Functionally it is a bipartite assembly platform: HEAT repeats 1–10 form the concave surface that recruits one of ~26 regulatory B-family subunits, while repeats 11–15 dock the catalytic C subunit, together forming the PP2A heterotrimer. Critically the solenoid is not a rigid ruler: inter-repeat hinge angles — concentrated around repeats 5–7 and the B/C-subunit interface — flex between 'open'/extended and 'clamped'/compact states depending on which subunits are bound. PR65 is a spring, not a fixed curve.
❌ Why AlphaFold gets it wrong
This is a confidence paradox. Mean pLDDT 96.8 says AlphaFold is essentially certain of the local geometry of every HEAT-repeat helix pair — reasonable, because the repeat unit is a simple, well-constrained motif. Yet TM-score 0.293, Cα-RMSD 15.0 Å, lDDT 0.311 and a 0.99 confidently-wrong-residue fraction show the assembled molecule is grossly wrong in overall shape against the deposited PR65 subunit of 8UWB.
It is a curvature/hinge failure, not a fold failure: AlphaFold locks onto one static superhelical bend, but tiny per-repeat angular errors compound multiplicatively across 15 tandem repeats into tens of Ångströms of end-to-end displacement — exactly what a global metric like TM-score punishes while the more local lDDT (0.31, not near-zero) partially forgives. The near-zero pLDDT↔lDDT correlation (−0.13) confirms confidence carries no diagnostic signal. The source paper had to solve an experimental PP2A trimer to guide AlphaFold2-Multimer rather than trust single-sequence prediction of the assembly.
🩺 Biology & disease
PP2A is one of the cell's two dominant serine/threonine phosphatases, an obligate A–B–C heterotrimer that dephosphorylates substrates across the cell cycle, Akt/PI3K and MYC signalling, apoptosis and the DNA-damage response — a major tumour suppressor. PPP2R1A is the sole scaffold onto which combinatorial exchange of B subunits builds an estimated ~100 distinct holoenzymes with different specificities and localisations.
PPP2R1A is recurrently mutated in endometrial and ovarian carcinomas, with hotspots (P179R, R183W/P, S256F, W257G/L/C, R258H) clustering in HEAT repeats 5–7 that disrupt B56-family binding without abolishing C-subunit docking — uncoupling substrate targeting from catalysis. Germline variants cause a neurodevelopmental disorder. PP2A reactivation by small-molecule activators (SMAPs) is an active oncology strategy exploiting exactly the scaffold flexibility this structure captures.
A single folded three-helix coiled-coil switch that toggles between a clamped-shut resting bundle and an extended active state — AlphaFold committed to one conformer, and the deposited structure is the other.
Worst deposition: 6YEL_A ·
drag to rotate · scroll to zoom · toggle the AlphaFold “ghost” to see what it predicted.
0.39
worst TM-score
78.86
mean pLDDT
0.75
FRAUD score
×1
structures
1/1
conf. wrong
2%
novelty
🧬 Structure & architecture
STIM1 is a type-I single-pass ER membrane protein: an N-terminal luminal EF-hand/SAM domain senses ER Ca²⁺, followed by one transmembrane helix, and a cytosolic region of stacked coiled-coil segments — CC1, then the CC2–CC3 pair that folds into the CAD/SOAR module that ultimately traps and gates Orai1. CC1 (~238–343) is not a simple two-stranded coiled coil but a compact, antiparallel three-helix bundle held together by a hydrophobic interhelical seam.
In the resting cell this bundle folds back on itself so that CC1α3 clamps CAD/SOAR intramolecularly, holding STIM1 'tight' and autoinhibited. Store depletion collapses the luminal Ca²⁺ signal, propagates across the membrane and releases the clamp, letting CC1 and the whole C-terminus adopt an 'extended' conformation that frees CAD/SOAR to engage Orai1. CC1 is a genuinely bistable folded module — two legitimate, functionally distinct tertiary arrangements of the same helical content.
❌ Why AlphaFold gets it wrong
This is emphatically not disorder or amyloid: SS-Q3 of 69.6% confirms AlphaFold gets the local chemistry right — CC1 is helical, and the model largely agrees on where the helices are. What collapses is tertiary packing: lDDT 0.341, TM-score 0.389 and Cα-RMSD 35.8 Å — the signature of correct secondary structure wrapped around the wrong global arrangement.
Consistent with AlphaFold's known bias toward continuous, extended coiled-coil geometry for heptad-rich sequences, it most likely predicted CC1 as a single elongated helix or open hairpin — the 'extended', CAD-releasing conformation — while 6YEL captures the compact, folded-back three-helix bundle of the autoinhibited state (or vice versa). A single-sequence predictor trained to output one static structure has no mechanism to represent a switch with two functionally required minima; it commits to one at pLDDT 78.9, leaving 84% of residues confidently wrong. The fold class is correct; the conformer is not.
🩺 Biology & disease
STIM1 is the ER-resident Ca²⁺ sensor that initiates store-operated Ca²⁺ entry (SOCE). Ca²⁺ loss from the EF-SAM domain triggers luminal unfolding and oligomerisation, redistribution to ER–plasma-membrane junctions, and release of the CC1 clamp so that CAD/SOAR can capture and gate Orai1 CRAC channels — driving Ca²⁺ influx essential for T-cell and mast-cell activation, muscle excitation–contraction coupling and platelet function.
The CC1α1–α2 contacts set the activation threshold. Gain-of-function mutations such as R304W disrupt these contacts, elongate the helix and destabilise the tight resting bundle, producing constitutive CAD exposure — underlying Stormorken syndrome and tubular aggregate myopathy. Loss-of-function STIM1 mutations abolish CRAC activity, causing CRAC-channelopathy immunodeficiency. CC1 is the structural fulcrum between healthy and diseased Ca²⁺ signalling.
Activator of Hsp90 ATPase 1 (Aha1) AHSA1
Right domains, wrong inter-domain pose (flexible linker)globularflexible linkerco-chaperoneO95433 ↗
Two well-folded domains on a flexible tether — AlphaFold committed to a single, plausible relative pose of a conformationally dynamic co-chaperone, and the deposited solution structure sampled a different one.
Worst deposition: 7DME_A ·
drag to rotate · scroll to zoom · toggle the AlphaFold “ghost” to see what it predicted.
0.43
worst TM-score
83.34
mean pLDDT
0.82
FRAUD score
×4
structures
1/4
conf. wrong
0%
novelty
🧬 Structure & architecture
Human Aha1 (338 aa) is built from two compact globular domains: an N-terminal domain (NTD, ~1–156, a Rossmann-like α/β fold) and a C-terminal domain (CTD, ~157–338, a curved β-sandwich), joined by a linker that is only partly ordered even when Aha1 is docked on its partner. Each domain individually folds robustly and is well characterised in isolation.
In the extended, Hsp90-bound state the NTD engages the Hsp90 middle domain and the CTD packs against the N-terminal ATPase domains, stabilising the closed, hydrolysis-competent state in an asymmetric 1:2 stoichiometry. But this domain-bridging arrangement is an induced state: free Aha1 in solution is intrinsically dynamic, its NTD and CTD tumbling largely independently and sampling a broad distribution of inter-domain orientations. Aha1 behaves less like one rigid two-domain protein and more like two folded modules loosely constrained by a flexible hinge.
❌ Why AlphaFold gets it wrong
The numbers tell an intra-domain-correct, inter-domain-wrong story. lDDT is a moderate 0.748 and SS-Q3 is 74.3% — both domains fold essentially correctly, far better than genuinely misfolded targets. But TM-score collapses to 0.432 with a 23.4 Å Cα-RMSD, and 85% of residues are confidently wrong despite mean pLDDT 83.3.
AlphaFold has built two accurate local folds, then bolted them together at one specific, self-consistent NTD–CTD angle and buried the joint in a high-pLDDT interface — exactly the behaviour its confidence metric should flag but does not (pLDDT↔lDDT correlation only +0.33), because the network has no way to represent an ensemble. The NMR solution structure (7DME) instead reflects the flexible, loosely coupled reality: whichever discrete pose AlphaFold picked, the linker's real freedom put the deposited orientation far from it. A textbook flexible-linker failure — the pieces are right, the assembly instruction is wrong.
🩺 Biology & disease
Aha1 is the strongest known stimulator of Hsp90's intrinsically slow ATPase, accelerating the conformational cycle required to mature a broad swath of Hsp90 clients — kinases, steroid hormone receptors and disease-relevant misfolded proteins. It binds Hsp90 asymmetrically (one Aha1 per dimer suffices), with its NTD engaging the Hsp90 middle domain and its CTD stabilising the closed N-domain interface, and also shows independent holdase-like chaperone activity toward stress-denatured proteins.
The partnership has direct disease relevance: Aha1 drives maturation of ΔF508-CFTR, and reducing Aha1 improves ΔF508-CFTR trafficking — a strategy of interest in cystic fibrosis; Aha1 also promotes accumulation of pathological tau, implicating it in tauopathies. Given Hsp90's centrality to oncogenic client maturation, the Aha1–Hsp90 interface is an actively pursued target for allosteric modulation in cancer and proteostasis disease.
Each WW module of WWP2 is a textbook three-stranded β-sheet AlphaFold folds correctly in isolation — but the deposited chain has no single fixed inter-module geometry, and AlphaFold guessed one anyway; tellingly, at only pLDDT 74.5, it was honestly unsure.
Worst deposition: 6RSS_A ·
drag to rotate · scroll to zoom · toggle the AlphaFold “ghost” to see what it predicted.
0.46
worst TM-score
74.50
mean pLDDT
0.71
FRAUD score
×7
structures
1/7
conf. wrong
8%
novelty
🧬 Structure & architecture
Full-length WWP2 (870 aa) has the canonical NEDD4-family layout: an N-terminal C2 domain, four WW domains (WW1–WW4) that read PPxY/LPxY motifs in substrates, and a C-terminal HECT domain that forms the E3~ubiquitin thioester. Each WW domain is one of the smallest autonomously folding units known — ~35–40 residues, a three-stranded antiparallel β-sheet stabilised by two invariant tryptophans packing its hydrophobic face.
In the resting enzyme WWP2 is autoinhibited: the 2,3-linker wraps onto the HECT domain, occluding its allosteric ubiquitin site — relieved by ubiquitin binding, multivalent WW engagement by adaptors such as Ndfip1, or linker phosphorylation. Separately, isoform-specific pairing of WW3 and WW4 recognises PY-motif partners (e.g. Smad7) with an inter-domain geometry that is not fixed: it shifts with ligand identity and phosphorylation state.
❌ Why AlphaFold gets it wrong
The chemistry is trivial: a two-Trp, three-strand β-sheet is exactly the small, high-contact-order motif AlphaFold nails routinely. Yet mean pLDDT here is only 74.5 — the model is honestly hedging — and everything downstream is bad: TM-score 0.46, lDDT 0.503, Cα-RMSD 21.9 Å, SS-Q3 63%, 73% of residues confidently wrong, and a pLDDT↔lDDT correlation of just +0.15.
The 109-residue deposited chain pairs the native WW fold with a second, unrelated folded module (an NMR solubility-tag partner) via a short flexible tether — structurally the same problem as native WW3–WW4 tandems having no single relative orientation. Given one sequence and no knowledge of which pose (or which fusion) was actually captured, AlphaFold predicted a specific, wrong relative packing — a linker/inter-module geometry failure layered on top of individually correct small folds, not a misfolded core. Here the honest thing is that its confidence, uniquely low among these examples, half-admits the uncertainty.
🩺 Biology & disease
WWP2 is a HECT-family E3 ubiquitin ligase of the NEDD4-like clan, using its WW domains as substrate-recognition modules that dock PY-motif partners for K48- or K63-linked ubiquitination. Validated substrates include SMAD7 and SMAD2/3 (TGF-β pathway), PTEN, TP53 and the pluripotency factor OCT4, alongside roles touching chromatin regulators and autophagy machinery.
Because WWP2 arises as multiple splice isoforms with different WW-domain complements, the same locus produces enzymes with opposing substrate preferences — WWP2-C preferentially degrades the TGF-β inhibitor SMAD7 and promotes EMT/metastasis, while WWP2-N targets SMAD2/3. Autoinhibition via the 2,3-linker keeps basal activity low until relieved by ubiquitin binding or phosphorylation. WWP2 dysregulation and isoform-ratio shifts are implicated in cancers (SMAD/TGF-β and PTEN/p53 axes), osteoblast and cardiac remodelling, and stem-cell biology via OCT4 turnover.
The dramatic failures — aggregation, disorder & context
The headline catches: whole-fold catastrophes where AlphaFold's confident native prediction and the deposited coordinates are essentially unrelated — a novel membrane amyloid, a disordered monomer forced into a fibril, a folded tetramer that amyloids, and a multidomain zymogen captured in the wrong biological context.
AlphaFold built TMEM106B a tidy, glycosylated FN3 β-sandwich at pLDDT 94 — the real deposited structure is a self-templating cross-β amyloid ladder that forms in essentially every aged human brain, and AlphaFold got every one of the ~135 ordered residues wrong.
Worst deposition: 7U14_A ·
drag to rotate · scroll to zoom · toggle the AlphaFold “ghost” to see what it predicted.
0.16
worst TM-score
94.39
mean pLDDT
0.87
FRAUD score
×29
structures
27/29
conf. wrong
100%
novelty
🧬 Structure & architecture
TMEM106B is a type-II single-pass lysosomal membrane protein: a short, intrinsically disordered cytoplasmic N-terminus (~1–96), a single TM helix (~97–120), and a luminal C-terminal domain (CTD, ~118–274). The monomeric CTD's native fold, solved by X-ray crystallography (PDB 8B7D), is a canonical fibronectin type III (FN3) β-sandwich — a seven-stranded Ig-like fold crowned by a short α-helix and pinned by a disulfide.
In disease and in aged brain the same CTD polymerises homotypically: residues ~120–254 form the ordered fibril core, adopting an entirely different five-layered cross-β architecture of 17–19 short β-strands per rung, stacked in-register at ~4.8 Å rise. Three N-glycans and a C214–C253 disulfide are built into the ordered core — unusual for amyloid. This is a polymer fold, not a monomer fold; AlphaFold predicts only the latter.
❌ Why AlphaFold gets it wrong
AlphaFold returns the native monomeric FN3 fold at mean pLDDT 94.4 — high confidence, because the sequence has strong coevolutionary support for a folded globular domain. But the deposited structure (7U14 chain A) is the amyloid cross-β core, and the two topologies are essentially unrelated: TM-score 0.157 (near-random), Cα-RMSD 27.1 Å, GDT_TS 0.56, and even secondary-structure identity collapses to Q3 52.6%.
The calibration failure is total: the pLDDT↔lDDT correlation is 0.03 (confidence carries no information about local accuracy), 72% of residues are over-confident by PAE, and 100% are confidently wrong. The root cause is structural: amyloid is a non-native, self-templated fold stabilised by intermolecular stacking across thousands of copies — a cooperative polymer state that cannot be inferred from a single chain. The sequence is also 100% novel to AlphaFold's training cutoff, so no template steers it off its globular-fold prior.
🩺 Biology & disease
TMEM106B is a broadly-expressed lysosomal membrane glycoprotein regulating lysosome size, pH, positioning and trafficking. It was the first GWAS risk/modifier locus for FTLD-TDP, acting most strongly in progranulin (GRN) mutation carriers, where the risk haplotype accelerates onset — largely via a linked coding variant, T185S, that changes glycosylation and turnover.
The 2022 discovery (reported near-simultaneously in Cell, Nature and Acta Neuropathologica) that the CTD itself forms amyloid fibrils — independent of and distinct from TDP-43, tau and α-synuclein — reframed TMEM106B as an amyloidogenic species in its own right. These fibrils accumulate across a broad range of neurodegenerative diseases and, strikingly, in cognitively-normal aged brain in an age-dependent manner.
AlphaFold confidently predicted a fold for a protein that, in reality, has no single fold to predict — and the 208 deposited structures that expose this are almost entirely amyloid fibrils, a self-templating cross-β architecture that only exists once thousands of copies stack together.
Worst deposition: 9A1Q_A ·
drag to rotate · scroll to zoom · toggle the AlphaFold “ghost” to see what it predicted.
0.14
worst TM-score
75.18
mean pLDDT
0.71
FRAUD score
×211
structures
207/211
conf. wrong
0%
novelty
🧬 Structure & architecture
Alpha-synuclein is a 140-residue intrinsically disordered protein with no stable tertiary structure as a soluble monomer. It has three regions: an amphipathic N-terminus (1–60) of imperfect KTKEGV repeats that folds into a helix on membrane contact; the hydrophobic NAC domain (~61–95), the aggregation-nucleating core; and an acidic, unstructured C-terminus (96–140) that chaperones against aggregation.
The deposited structures are full-length fibrils solved by cryo-EM: two protofilaments intertwined about a 2₁ screw axis, each a serpentine, Greek-key-like arrangement of β-strands packing a hydrophobic core over residues ~37–99. This cross-β fold is polymorphic — Parkinson's/DLB filaments differ from the multiple-system-atrophy polymorphs — and is templated by intermolecular stacking, not by the monomer's own energy landscape.
❌ Why AlphaFold gets it wrong
On the worst case (9A1Q chain A): TM-score 0.138, Cα-RMSD 36.3 Å, lDDT 0.18, GDT_TS 1.6, SS-Q3 35% — essentially no fold correspondence. Expected, since AlphaFold has no cross-β amyloid mode in its training distribution, and nothing about the sequence alone specifies which disease-associated protofilament fold it will adopt.
More diagnostically, mean pLDDT 75.2 with a pLDDT↔lDDT correlation of −0.29 shows AlphaFold's confidence is actively anti-correlated with accuracy — where the model is more confident it is measurably wrong more often, likely because it partially recognises local propensities (helical N-terminus, extended NAC) without any signal for polymorph-specific packing. This is the archetypal disordered-monomer-forced-into-amyloid failure: 206 of 208 depositions confidently wrong, the highest deposition count in the set.
🩺 Biology & disease
Alpha-synuclein is a presynaptic protein that modulates SNARE-complex assembly, vesicle clustering and neurotransmitter release, binding curved acidic membranes via its N-terminal amphipathic helix. Its aggregation into fibrils in Lewy bodies defines the synucleinopathies: Parkinson's disease, dementia with Lewy bodies, and multiple system atrophy (where fibrils accumulate in oligodendrocytes).
Familial PD mutations (A30P, E46K, H50Q, G51D, A53T) cluster in the N-terminal/NAC region and alter membrane binding and polymorph selection; SNCA locus multiplication causes disease by dosage. Pathology spreads by prion-like templated seeding, and distinct fibril 'strains' correlate with distinct clinical syndromes — a proposed basis for seed-amplification biomarkers.
AlphaFold calls the native TTR fold at pLDDT 98 — statistically indistinguishable from certainty — while the deposited cardiac amyloid fibril scores TM 0.21 against that prediction, a 22.8 Å Cα-RMSD collapse that makes this one of the starkest confidently-wrong cases in the dataset.
Worst deposition: 9BZS_A ·
drag to rotate · scroll to zoom · toggle the AlphaFold “ghost” to see what it predicted.
0.21
worst TM-score
98.01
mean pLDDT
0.95
FRAUD score
×160
structures
34/160
conf. wrong
0%
novelty
🧬 Structure & architecture
The TTR monomer is a 127-residue subunit folded as an eight-stranded β-sandwich — two four-stranded antiparallel sheets in a Greek-key, immunoglobulin-like topology with a short α-helix between strands E and F. Monomers dimerise edge-to-edge and two dimers pack back-to-back into the physiological homotetramer, forming two funnel-shaped thyroxine-binding sites. This is exactly what AlphaFold predicts, at near-maximal confidence.
The amyloid state is a different molecule. Tetramer dissociation and partial monomer unfolding precede assembly into cross-β fibrils: cryo-EM resolves a single protofilament built from an N-terminal segment (~Pro11–Lys35) and a C-terminal segment (~Gly57–Thr123) packed into a compact serpentine β-arch, connected by a disordered linker, stacking at ~4.8 Å rise. Essentially none of the native sandwich topology survives.
❌ Why AlphaFold gets it wrong
Every metric tells the same story. AlphaFold's mean pLDDT 98.0 reflects extreme confidence in the native β-sandwich — unsurprising, since TTR has 0% sequence novelty and sits squarely in the training set. But the deposited structure (9BZS chain A) is the amyloid fibril: TM-score 0.214, Cα-RMSD 22.8 Å, lDDT 0.49, GDT_TS 0.27, SS-Q3 56.5%. The confidently-wrong-residue fraction is 1.0, and 75% of errors fall in PAE-overconfident territory; FRAUD 0.95.
The mechanistic point is the cleanest illustration of the whole thesis: pLDDT measures AlphaFold's confidence in the fold it knows — the evolutionarily-conserved native state — not whether a given deposited coordinate set represents that state. Amyloid is a non-native, kinetically-accessed aggregation product; nothing in the sequence-to-native-structure mapping AlphaFold learned encodes it. This is a wrong-STATE failure, not an unknown-sequence one.
🩺 Biology & disease
TTR is a homotetrameric transport protein made mainly by the liver (plasma) and choroid plexus (CSF). It carries thyroxine (T4) in the tetramer channel and retinol indirectly, via retinol-binding protein 4. Amyloidogenesis is rate-limited by tetramer dissociation to monomer, which partially unfolds and reassembles into fibrils that deposit in nerve and cardiac tissue.
Two diseases result: wild-type ATTR (age-related cardiac amyloidosis) and hereditary ATTRv from >130 destabilising mutations (classically V30M polyneuropathy). Uniquely among amyloid diseases, it is genuinely druggable — tetramer kinetic stabilisers (tafamidis, diflunisal, acoramidis) lock the tetramer at the T4 sites, and gene-silencing agents (patisiran, vutrisiran) suppress hepatic production. The 2024 cryo-EM work behind 9BZS found V30M fibrils from heart and nerve of the same patient are near-identical.
Coagulation factor X F10
Wrong context — fragment / complex / PTMnot amyloidmultidomaindruggableP00742 ↗
AlphaFold nails the modular zymogen it was trained to see, but the worst-scoring deposition isn't Factor X folded at all — it's a 39-residue stretch of the EGF1 domain, unfolded and threaded through the active site of a partner hydroxylase: a failure of context, not of fold prediction.
Worst deposition: 6Q9F_B ·
drag to rotate · scroll to zoom · toggle the AlphaFold “ghost” to see what it predicted.
0.13
worst TM-score
93.49
mean pLDDT
0.37
FRAUD score
×47
structures
26/47
conf. wrong
0%
novelty
🧬 Structure & architecture
Full-length Factor X is a vitamin-K-dependent zymogen built from an N-terminal Gla domain (11 γ-carboxyglutamates coordinating Ca²⁺ for membrane docking), two tandem EGF-like domains, and a C-terminal chymotrypsin-fold serine protease domain (catalytic triad His236–Asp282–Ser379). Light and heavy chains are disulfide-linked; a 52-residue activation peptide is excised on activation to factor Xa. AlphaFold predicts this compact multidomain architecture with high confidence.
The deposited PDB corpus fragments that picture: isolated Gla domains, isolated protease domains, activated Xa, Xa in prothrombinase/tenase assemblies — and, in the worst case here (6Q9F chain B), a synthetic 39-mer peptide spanning the EGF1 β-hydroxylation site, captured extended through the active-site channel of aspartyl/asparaginyl β-hydroxylase (AspH), not folded as an EGF module at all.
❌ Why AlphaFold gets it wrong
This is a wrong-context failure, not re-folding-into-amyloid, and the numbers say so. FRAUD 0.37 and Cα-RMSD 6.6 Å are moderate — nowhere near the ~25 Å amyloid catastrophes — and SS-Q3 70% shows partial secondary-structure agreement survives. Yet TM-score collapses to 0.127: a 39-residue linear fragment threaded through AspH's channel has no compact fold to superpose against AlphaFold's EGF1-in-context prediction.
The pLDDT↔lDDT correlation of −0.59 is the tell: AlphaFold assigns high local confidence (mean pLDDT 93.5) to a folded domain context that isn't present in the crystal, so confidence and correctness move in opposite directions. AlphaFold models the isolated, intact, canonical protein; experiments capture it activated, cleaved, fragmented or bound — states that, given one sequence, it cannot know. It has no way to anticipate a residue will later be excised as substrate for another enzyme.
🩺 Biology & disease
Factor X is a vitamin-K-dependent plasma zymogen made in the liver. It is activated by two convergent tenase complexes — the extrinsic tissue-factor/FVIIa and the intrinsic FIXa/FVIIIa (deficient in haemophilia B and A). Activated factor Xa, with cofactor Va, Ca²⁺ and phospholipid, forms prothrombinase, the complex that converts prothrombin to thrombin — the central amplification node of coagulation.
That makes Xa one of the most validated anticoagulant targets: the direct oral inhibitors rivaroxaban, apixaban and edoxaban bind the protease S1/S4 pockets and are first-line for atrial fibrillation and venous thromboembolism. Separately, AspH hydroxylates a conserved Asp/Asn in Factor X's EGF1 domain using Fe(II)/2-oxoglutarate chemistry — an oncofetal antigen of interest in cancer immunotherapy, which motivated this exact structure series.
