Live Stats, next update: Wed 29 Jul
Human PDBs Analysed
Confidently Wrong
Novel + Confidently Wrong
DB size
Visitors
Full statistics →
New PDB Depositions vs. Their Blind AlphaFold Predictions — A Running Test of “Is Folding Solved?”

Welcome to AlphaFraud

AlphaFraud watches the PDB for freshly-deposited human structures and pits each against its blind AlphaFold prediction — a continuous, public test of whether protein folding is really “solved”. Every week it surfaces the structures AlphaFold got most confidently wrong: sure of itself, and still missing.

82,884
structures analysed
1,945
confidently wrong
2.3%
confidently-wrong rate
431
novel + confidently wrong
0.5%
novel + wrong rate
457
release weeks

Explore the site

🧬

All structures

The cumulative dashboard — every analysed human structure on the pLDDT-vs-agreement “fraud quadrant”, with metric histograms and a weekly trend.

💡 You'll learn: Spot the confidently-wrong cluster: high AlphaFold confidence, low real agreement.
Open All structures →
🏆

Worst offenders

The all-time leaderboard of AlphaFold’s biggest misses, grouped by protein with a ×N repeat count and a deviation-coloured ribbon each.

💡 You'll learn: Which proteins AlphaFold fails on again and again — amyloids top the list.
Open Worst offenders →
🔬

Examples

Hand-authored deep-dives plus an auto “example of the week”, opening with a positive control where AlphaFold nailed a novel structure.

💡 You'll learn: Why AlphaFold gets some folds right and mangles others — explored in interactive 3D.
Open Examples →
🧪

Calculate

Type any human PDB ID and score it on demand — the full pipeline runs live and returns the same panel, with no AI-written text.

💡 You'll learn: Exactly how AlphaFold did on a structure you care about.
Open Calculate →
📈

Analysis

Where the failures cluster: CATH/SCOP fold-family enrichment, sequence clustering and per-superfamily blind-spot scorecards.

💡 You'll learn: Which fold classes AlphaFold systematically struggles with.
Open Analysis →
🗄️

Archive

Every PDB release week since the 2018 cutoff, browsable at weekly granularity and linked to each week’s structures.

💡 You'll learn: Track the confidently-wrong rate week by week.
Open Archive →