Elucidating enzymatic mechanisms is essential for improving the prediction of enzyme function and designing new biocatalysts. MechFind is an optimization-based approach developed to predict plausible reaction mechanisms solely from the substrate and product of enzymatically catalyzed reactions. Using a mixed integer linear programming (MILP) framework, MechFind identifies mechanisms as sequences of mechanistic steps involving bond formations, cleavages, and electron transfers, ensuring mass and charge balance at each step. A machine learning model is then used to score the proposed mechanisms. A holdout validation procedure using an 80/20 train-test split correctly identified the mechanisms among the top three predictions for 74% of the test samples. MechFind was applied to more than 20,000 biochemical reactions from the MetaNetX database, identifying the three most probable mechanisms. MechFind provides a structured and quantitative framework for predicting detailed enzyme mechanisms, integrating a MILP-based search with machine learning based score for ranking. MechFind can be essential to guide the design of enzymes for novel reactions. MechFind is accessible on GitHub at https://github.com/maranasgroup/MechFind.
