One water molecule swings AI enzyme design hit rate from 15% to 88%, Nature paper shows
bravo_abad · x · 2026-10-15
A Nature paper from David Baker's group (Chen et al., Oct 14) on computational design of metalloproteases shows a single overlooked water molecule can mislead AI enzyme design.
- The designed enzymes use zinc to activate a water molecule for peptide cleavage; the water must sit exactly right next to the peptide
- With RFdiffusion2-MI for generation and AlphaFold3 for structure checking, active and inactive designs looked identical without the reacting water in the prediction; including it revealed the inactive design's peptide shifting out of position
- Experimentally: 15% of designs selected with zinc-only models passed the activity test vs 88% with zinc–water models (hit = >10% substrate cleavage in 12 hours)
Takeaway: explicitly modeling catalytic water in AI protein design multiplies hit rates.
Related event: Baker Lab Publishes Two Nature Papers on De Novo AI-Designed Proteases(6 posts)→
More from Research
- Interpretability debate: 'features' as fundamental neural network units may be wrong — MoonL88537 · 2026-10-16
- FOMO-Dex adds force-aware compliance to dexterous teleoperation for contact-rich robot tasks — HaozhiQ · 2026-10-16
- New mech interp hypothesis: neural networks generalize from the very first training step — burny_tech · 2026-10-16
- Princeton launches research initiative on keeping autonomous AI systems aligned with human intent — LakeBrenden · 2026-10-16
- OrganicHAR at UbiComp 2026: Discovering Activities Sensors Can Actually Distinguish — mynkgoel · 2026-10-16
- Dev critiques compiling fast programs into slow neural nets: fun, but not useful — ctjlewis · 2026-10-16