Machine-Learning-Guided Library Design Cycle for Directed Evolution of Enzymes: The Effects of Training Data Composition on Sequence Space Exploration
Problem
Directed evolution of enzymes is slow, and it was unclear how machine-learning training data composition affects guidance of sequence-space exploration.
Why it matters
Faster, ML-guided enzyme improvement cuts cost and time for protein engineering in industry and medicine.
What the AI did
ML predicted variant libraries for Sortase A over two rounds; predicted variants were experimentally tested and fed back as training data, with and without a known highly positive variant (5M).
Results
Enzyme activity 2.2–2.5 times higher in both series; improvements comparable with and without the known positive variant.
Impact
Shows ML can guide directed evolution effectively regardless of a known highly positive variant in training data.
Record
| Venue | ACS Catalysis |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Yutaka Saitō, Misaki Oikawa, Takumi Sato, Hikaru Nakazawa, Tomoyuki Ito, Tomoshi Kameda, Koji Tsuda, Mitsuo Umetsu |
| Citations | 70 (OpenAlex, at import time) |
| References | 51 |
| Model or method | not stated in source |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1021/acscatal.1c03753 |
| Open access | hybrid |
| Keywords | machine learning, directed evolution, Sortase A, protein engineering, variant library |
Importance rating
real but narrower advance
Single-enzyme (Sortase A) lab study on ML-guided directed evolution; abstract frames ML as attractive for protein engineering but gives no scale of impact beyond this narrow demonstration.