This project proposes to develop computational techniques for drug discovery based on accurate prediction of ligand-protein interactions using a novel polarizable potential energy function. The ability to predict binding affinity and specificity of a ligand to a macromolecular target is the key to structure-based drug discovery in silico. Currently available approaches are seriously limited by the quality of the underlying potential energy function that describes molecular interactions. Our goal is to achieve chemical accuracy in predicting ligand- protein binding by combining accurate physical based potentials with efficient computational algorithms. We will utilize and optimize a novel potential energy function with explicit polarization and atomic multipole electrostatics for predicting ligand-protein binding. We will develop a range of computational models and programs based on this polarizable potential function for lead discovery, including a drug-like fragment library prepared with atomic polarizability and multipole parameters;a program to automate ligand parameterization;a generalized Kirkwood implicit solvation model for polarizable atomic multipoles that can be applied to virtual screening of ligand library;a multiscale hybrid model that combines polarizable and fixed charge potential model for ligand-protein binding calculations;and parallel simulation programs to compute ligand-protein binding free energy with molecular dynamics. We will focus on metalloproteins that present a particular challenge for current approaches. Validation will be made via direct comparison to experimental binding affinities and structures of well studied systems including trypsin and matrix metalloproteinases. Our ultimate objectives are to develop efficient computational technology and software tools that enable computational drug discovery with chemical accuracy, and at the same time to advance our understanding of physical and chemical principles in molecular recognition. Computational models and algorithms developed in this work will also empower the broader molecular modeling community with a next-generation research tool. The resulting programs and parameters will be compatible with the existing drug discovery and molecular modeling software packages including TINKER, AMBER and AutoDock while documented source codes will be made freely available to the public.
Showing the most recent 10 out of 34 publications
