Breaking the Scale Barrier in Biocatalyst Development: AI-Powered Enzyme Engineering for Minimal Variants and Commercial Efficiency

A sustainable biocatalytic pathway for an industrially relevant amine was developed by integrating enzyme engineering with green chemistry principles. Conventional optimization of R-transaminases for pharmaceutical synthesis typically requires screening thousands of variants, often leading to destabilizing multimutation combinations that compromise solubility and catalytic efficiency. Here, we report an AI-driven 6D-grid protein-engineering framework that integrates interaction energy descriptors, solvent effects, and a data set of 1.39 million structural fragments to predict and validate productive substitutions. Using this approach, five AI-prioritized variants, each containing nine mutations, were evaluated and exhibited high solubility and catalytic stability at the 7 L fermentation scale. The engineered enzyme converted a prochiral ketone to sitagliptin with 51% conversion at 20% enzyme load and >99% enantiomeric excess, further reaching 89% conversion upon scale-up. To improve process sustainability, DMSO was replaced with a biodegradable ethanol/PEG-400 cosolvent system, where the engineered transaminase achieved >80% conversion under higher enzyme load, meeting industrial benchmarks. This work demonstrates how data-driven intelligence can minimize experimental screening while delivering scalable, environmentally responsible biocatalysis.