Researchers have made a breakthrough in creating novel serine hydrolase enzymes capable of efficiently catalyzing ester hydrolysis, using a computational approach that combines machine learning and protein engineering.
The study, published in Science, employs a new machine learning model called PLACER (Protein-Ligand Atomistic Conformational Ensemble Reproduction) to predict the active site conformations of designed enzymes. This allows for precise design and prediction of not just the overall protein fold but also the conformations of individual sidechains.
The team began by designing proteins with complex catalytic sites using the established generative AI framework, RFdiffusion. They then evaluated the active site conformation with PLACER and successfully created serine hydrolases that efficiently catalyze ester hydrolysis.
Initially, the “designed enzymes could only perform the first half of the reaction mechanism before becoming inactivated,” co-lead author Anna Lauko explained. The team redesigned simplified versions containing three of the five catalytic groups in the natural enzymes and eventually created a more complex design containing all five catalytic groups found in the native serine hydrolases.
Through experimental characterization, the team found that the novel enzymes retained folds unique from natural serine hydrolases with high catalytic efficiencies. The study’s findings highlight the potential of computational tools in enzyme engineering and their application in creating biocatalysts for industrial and pharmaceutical purposes.
Source: https://www.genengnews.com/topics/artificial-intelligence/ai-driven-protein-design-produces-enzyme-that-mimics-natural-hydrolase-activity