Chain of Structures: reasoning in protein space hits SOTA on 5 of 6 enzyme design metrics

量子位 · wechat · 2026-10-15

• Scinetics, with researchers from MIT, Harvard, Cambridge and HKUST, proposes representing scientific objects (sequences, structures, images, omics) as tokens that alternate with language during reasoning.
• Chain of Structures (COS) applies this to enzyme design in four rounds: catalytic core, active site, global backbone, then sequence and side chains, with each round producing a real structural state.
• COS achieves new SOTA on 5 of 6 benchmark metrics: EC3 at 57.0% (vs CodeFP's 55.3%), pLDDT 79.3, TM-score 0.729; ablations show both the reasoning chain and the continuous interface are needed.
• The companion Perspective outlines ScienceRSI, a recursive self-improvement loop where verified scientific trajectories train the next model generation, targeting Scientific SuperIntelligence with a 100-300B parameter exploration window.

Original post →

More from Research

• Domingos Cites New Result Claiming RL Is 'Basically Useless' vs Supervised Learning — pmddomingos · 2026-10-23
• Wolfram Opens Rule 30 Prizes to AIs: $30,000 for Three Open Problems — cephaloform · 2026-10-23
• Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23
• EMNLP paper: diffusion LMs internally encode latent denoising timestep representations — gsarti_ · 2026-10-23
• KU Leuven and VIB open two Group Leader professorships in AI for Biology, start 2027 — jpillowtime · 2026-10-23
• Subspace Tuning curbs LLM catastrophic forgetting by freezing principal singular directions — scottniekum · 2026-10-23

Research channel →