Testing Claude Fable at direct protein 3D coordinate prediction: TM-scores of just 0.19-0.25

LeoTZ03 · x · 2026-11-18

Someone tested Claude Fable 5.2 via API, asking it to output C-alpha 3D coordinates directly from sequence for three proteins, then compared against crystal structures. The results are far from usable: TM-scores of 0.19-0.25, C-alpha RMSD of 10.5-14.5 Å, with secondary structure and fold topology clearly mismatched — a reminder that chat models emitting raw coordinates are nowhere near AlphaFold-class structure prediction.

Related event: Researcher tests whether Claude Fable can predict protein contact maps from sequence alone(12 posts)→

Original post →

More from Fun

Fun channel →