Methanol tolerance upgrading of Proteus mirabilis lipase by machine learning-assisted directed evolution

For many crucial industrial applications, enzyme-catalyzed processes take place in harsh organic solvent environments. However, it remains a challenging problem to improve enzyme stability in organic solvents. This study utilized the MLDE (machine learning-assisted directed evolution) protocol to improve the methanol tolerance of Proteus mirabilis lipase (PML). The machine learning (ML) models were trained based on 266 combinatorial mutants. Using top 3 in 22 regression models based on evaluation of tenfold cross-validation, the fitness landscape of the 8000 full-space combinatorial mutants was predicted. All mutants in the restricted library showed higher methanol tolerance, among which the methanol tolerance of G202N/K208G/G266S (NGS) was up to 13-fold compared with the wild-type. Molecular dynamics (MD) simulation showed that reconstructing of critical hydrogen bond network in the mutant region of NGS provides a more stable local structure. This compact structure may improve the methanol tolerance by preventing organic solvent molecules into the activity site and resisting structural destruction. This work provides a successful case of evolution guided by ML for higher organic solvent tolerance of enzyme, and may also be a reference for broad enzyme modifications.