Protein design must account for neighbors: 578 variants show similarity predicts function poorly
bravo_abad · x · 2026-09-25
Bülbül et al. argue protein domains aren't reusable Lego bricks — function depends on neighbors. Using ESM3, ProteinMPNN and EvoDiff, they generated new thiolation domains for NRPS while fixing neighboring domains, generating the missing sequence inside its local A–T–C context.
Key findings across 578 recombinant variants:
- Sequence similarity to the native domain was a poor predictor of function;
- The identity of the immediate downstream partner mattered far more — designs made for one class of downstream domain mostly failed when moved to the other.
The model isn't just finding "good sequences" — it's learning context-dependent fitness.
Related event: 578 Variants Show Protein Domains Must Be Designed in Context(2 posts)→
More from Research
- Visualizing Raw Activation Values Could Seed a Math Framework for Interpretability — brandon_xyzw · 2026-09-25
- NVIDIA Talks Cosmos 3: Simulation and Verifiable Rewards for Physical AI — liu_mingyu · 2026-09-25
- Open-source fly Matrix: 150,802-neuron connectome flies in your browser — mtizard · 2026-09-25
- IDS paper wins NeurIPS oral: AI writes formally verified distributed systems 200x faster — AccBalanced · 2026-09-25
- NeurIPS desk-rejects 5/5/5-scored paper over one co-author's policy violation — AccBalanced · 2026-09-25
- C5R builds an AI-run research facility in 12 weeks, launches SciUniverse benchmark — akilian · 2026-09-25