Machine learning-guided co-optimization of fitness and diversity facilitates combinatorial library design in enzyme engineering
Problem
Designing combinatorial mutation libraries that balance fitness and diversity is difficult, especially for poorly characterized or unknown enzyme functions.
Why it matters
Engineered enzymes can enable new-to-nature reactions such as enantioselective C-B and C-Si bond formation, valuable for chemistry and industry.
What the AI did
MODIFY, an ML algorithm learning from natural protein sequences, infers evolutionarily plausible mutations, predicts enzyme fitness, and co-optimizes fitness and diversity of starting libraries for directed evolution.
Results
Outperformed state-of-the-art unsupervised methods in zero-shot fitness prediction in silico; engineered biocatalysts achieving enantioselective C-B and C-Si bond formation via new-to-nature carbene transfer, six mutations from previously developed enzymes.
Impact
Peer-reviewed demonstration of ML-guided directed evolution yielding functional new-to-nature biocatalysts.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Kerr Ding, Michael Chin, Yunlong Zhao, Wei Huang, Binh Khanh Mai, Huanan Wang, Peng Liu, Yang Yang et al. |
| Citations | 83 (OpenAlex, at import time) |
| References | 70 |
| Model or method | MODIFY |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-024-50698-y |
| Open access | gold |
| Keywords | enzyme engineering, machine learning, directed evolution, biocatalysis, zero-shot fitness |
Importance rating
significant — thousands to millions affected
Abstract states library design facilitates useful, poorly characterized enzyme functions beyond classic directed evolution; no scale stated.