AI model boosts enzyme thermal stability by up to 37°C without paired data
KevinKaichuang · x · 2026-11-07
A Chalmers University team (Zelezniak lab) posted a bioRxiv paper showing a model that stabilizes enzymes by up to +37°C while maintaining or increasing activity, without paired mesophilic/thermophilic training examples.
- Data pipeline: 51M metagenome sequences (JGI) and 203M UniProtKB entries filtered to 6.7M/11M non-redundant enzymes, organized into 4,471 EggNOG COG families (4,166 training, 10 held-out validation with 450k sequences)
- Method: a generator module plus identity/pID controllers, evolutionary-consistency and thermal-adaptation discriminators enforce family-consistent substitutions and thermal adaptation
- Cross-family learning transfers thermal adaptation between enzyme families
More from Research
- LLM agent re-architects models during training: algorithmic reasoning 55%→84%, AIME 64%→71% — lateinteraction · 2026-11-07
- UCLA's SidewalkBench: A Sim Benchmark for Sidewalk Robot Navigation, CoRL Spotlight — zhoubolei · 2026-11-07
- RareDiseaseBench: AI agents max out at 43.8% on 35 rare-disease diagnostic workups — kenbwork · 2026-11-07
- VIALS benchmark: frontier models flunk 161 life-sciences image interpretation tasks — niloofar_mire · 2026-11-07
- Sleep-time Compute paper cuts test-time compute ~5x by pre-thinking offline — RLanceMartin · 2026-11-07
- Cornell's Trainspotting: 12k 3D objects reconstructed from 311k found photos — Jimantha · 2026-11-07