ML Framework Evolves Enzymes to Access All Eight Stereoisomers
bravo_abad · x · 2026-08-25
Zhenyu Lu et al. published in Nature Synthesis a method using ML-guided protein engineering to evolve a single alcohol dehydrogenase into a complete set of stereocomplementary enzymes capable of producing all eight stereoisomers of alcohol products.
Technical Approach (BioStrucTag):
- Structure Branch: Represents the 3D catalytic pocket as voxelized ligand–enzyme complexes, allowing the model to learn structural signatures linked to stereochemical outcomes.
- Sequence Branch: Incorporates sequence info from a transformer-based protein language model.
- Synergy: The branches prioritize mutations without exhaustive screening, significantly shrinking the experimental search space.
Results: The resulting variants access all eight stereoisomers, with selectivities reaching 99:1 enantiomeric and diastereomeric ratios.
More from Research
- Open Source AI Eval Skills Guide Coding Agents — petergyang · 2026-08-25
- Critique of LoCoMo and LongMemEval: Seeking better memory benchmarks — LowDistribution3995 · 2026-08-25
- Standard Machines launches to teach AI advanced chip design via RL — ycombinator · 2026-08-25
- SWE-Bench ProMax Benchmark Targets Large-Scale Code Refactoring — rseroter · 2026-08-25
- Optima now supports benchmarking Qwen3.8 27B — ArtificialAnlys · 2026-08-25
- PhysCaP: Physics-Informed Agent for Robotic Manipulation — Chen-Yu Lin · 2026-08-25