249: Developing machine learning program for simulating ligand-enzyme docking

Abstract: Catalysis occurring in micellar environments serves to accelerate biochemical reactions. Serine proteases are essential in many biological processes across different organisms. The goal of this research is to develop a machine learning-based simulation framework capable of using enzyme structures, ligand docking, and substrate information to predict catalytic efficiency within micelles. This machine learning framework will be able to model enzyme-substrate interactions within micelles, enzyme catalysis within micellar environments, binding affinity, and reaction dynamics. A dataset of serine proteases structures from the Protein Data Bank was collected with a dataset of substrate molecules. Molecular docking was performed to identify binding poses within the enzyme’s active site. The model was trained using labeled derived data from experimental kinetics and computational simulations. Graph neural networks were also implemented to learn substrate features and predict catalytic efficiency. Computational simulations were used to perform molecular dynamic simulations of enzyme-ligand interactions within enzymes and to study how the enzyme environment influences transition state stabilization and binding affinity. Computational predictions were compared to available experimental data. The machine learning framework is currently still in development, with a large dataset of serine protease structures and substrate molecules already compiled. Molecular docking is being performed on molecules from the datasets to study binding poses, binding affinity, and reaction dynamics. Studying ligands and molecular structures through computational means such as docking may allow this framework to be helpful in both structure-based and ligand-based drug design.