Recalling homologs with Fable 5.2 for de novo protein coevolution: Top7 hits 35% of top-L contacts
sokrypton · x · 2026-11-20
The author used protein LM Fable 5.2 to "recall" homolog-like sequences for coevolution analysis of de novo proteins: Top7 reached 35% correct top-L contact pairs (Neff 1300, shuffled baseline 7%), while 6MRR only hit 9%. The model initially refused, citing no homologs, and was bypassed by prompting it to treat the query as natural. Performance remains below that on natural proteins.
More from Research
- Did Claude find 872 astrophysics paper errors? Physicists push back on false positives — jessyjli · 2026-11-20
- Implementing DEFLATE Compression From Scratch in C, Following RFC 1951 — tetsuoai · 2026-11-20
- A photonic chip plays Mario using a NEAT network trained for just 472 generations — CatAstro_Piyush · 2026-11-20
- Huawei Cuts Generative Recommendation Inference Latency up to 25x vs vLLM Without Custom Kernels — _reachsumit · 2026-11-20
- iWSP: LLM Agents Build Interpretable Neural Nets by Programming Weights Directly — kalyan_kpl · 2026-11-20
- Tencent's CONCORD-Rec Weights Each Modality by Reliability in Diffusion-Based Recommendation — _reachsumit · 2026-11-20