This video summarises the paper “Accelerated enzyme engineering by machine-learning guided cell-free expression,” published in Nature Communications (2025, PMID: 39833164).
Traditional directed evolution is slow. Each round of mutation and screening can take weeks, and often explores only a small corner of protein sequence space. This study introduces a new framework that combines cell-free gene expression with machine learning, enabling thousands of enzyme variants to be built, tested, and optimised within a day.
By training ML models on single-mutation fitness data, the team predicted multi-mutation enzyme variants with dramatically higher activity, up to 42-fold improvement, for nine pharmaceutical reactions catalysed by the amide synthetase McbA. Their approach bypasses cell transformation steps, reduces screening burden, and allows parallel evolution of multiple specialized biocatalysts from a single enzyme scaffold.
This work marks a step toward self-driving enzyme engineering, where high-throughput experimental data and simple computational models together accelerate biocatalyst design for sustainable chemistry and drug manufacturing.
source