AlphaFold3 and ESMFold2 fail on orphan proteins; Claude contact prediction still weak
sokrypton · x · 2026-11-20
sokrypton (of ESMFold fame) benchmarked contact prediction on orphan proteins with no MSAs: AlphaFold3 and ESMFold2 both fail on these sequences, and Claude models also predict contacts poorly beyond a few local beta-pairings.
He ran a careful reproducibility check — 3 runs across 3 Claude models — and found "Fable 5.2" performs better at contact prediction than other Claude models and has improved over Fable 5.1. Still, the plotted results show four orphan proteins with only 14%-28% of top-L pairs correct (long-range precision 0.01-0.05), far from practical. A rigorous negative-result report on LLMs for protein structure prediction.
More from Research
- Transformers PR streams MoE experts from SSD to run a 180B model on a 32GB Mac — jomic01 · 2026-11-20
- The Hundred-Page Machine Learning Book on sale for $14 (from $40) on Leanpub — burkov · 2026-11-20
- Image Models Now Ship Fine-tuned LLM Prompt Upsamplers — Why Not Fuse Them? — ostrisai · 2026-11-20
- PyTorch's ezyang on formalizing a trivial in-place kernel under PTX semantics — ezyang · 2026-11-20
- Scientist: Gemini 4 Neon designed innovative CAR receptors to kill cancer cells — DeryaTR_ · 2026-11-20
- New arXiv paper: Bregman power divergences turn KL queries into Euclidean NN in d+1 — FrnkNlsn · 2026-11-20