Enhancing Activity and Stability of Transaminase through Integrated Machine Learning, Rational Design, and Directed Evolution Approaches

Transaminases (ATAs) are promising biocatalysts for chiral amine synthesis but often suffer from limited activity and stability, particularly with non-natural substrates. This study integrates machine learning, rational design, and directed evolution to engineer Bacillus megaterium transaminase (BmATA) for synthesizing the Alzheimer’s drug precursor (S)-1-(3-methoxyphenyl)ethylamine. By employing machine learning algorithms with appropriate feature selection, we identified key mutations that enhanced catalytic properties while maintaining the structural stability. Starting from the wild-type BmATA with merely 4% conversion from 20 mM 3-methoxyacetophenone (1a), the initial engineering efforts yielded a mutant M6X with conversion rates of 95 and 8% from 20 and 50 mM substrates, respectively. Further optimization through disulfide bond design and directed evolution led to the development of the M12X2 mutant with a melting temperature of 77.6 °C, achieving a remarkable conversion rate of 92% from 50 mM 1a. These findings not only underscore the potential of combining computational and experimental approaches in ATA engineering but also highlight the effectiveness of M12X2 as a robust biocatalyst for chiral amine synthesis, paving the way for its future applications in pharmaceutical development.