Towards High-resolution Computational Approaches for Structure-based Drug Discovery

This dissertation describes new computational approaches at high resolution for practical structure-based drug discovery. It begins with a brief review of structure-based computational approaches for drug discovery in comparison with ligand-based ones, followed by a discussion of important applications in selecting drug-like compounds and predicting drug metabolites. Since three-dimensional target structures are crucial for structure-based drug discovery, a new methodology based on force fields for protein structure refinement was developed. This methodology employs the VSGB 2.0 energy model in combination with a robust protonation state assignment algorithm and efficient sampling strategies. High accuracy was obtained for predicting 2239 protein side chains and 115 14-20 residue loops. Given the precision and uniform robustness, this methodology is believed for the first time to be suitable to tackle practical problems in structure-based drug discovery. The VSGB 2.0 energy model was then applied in the development of a new accurate approach (IDSite) for predicting P450-mediated drug metabolism, a problem of great practical interest for drug discovery. IDSite is able to efficiently model induced-fit effects using flexible docking and constrained refinements. Sites of metabolism are determined based on the physical interactions between a P450 enzyme and the ligand. Preliminary tests with 56 compounds displayed both low false positive and low false negative rates, which demonstrate the high potential of IDSite to be used in metabolism tests for drug discovery. In conclusion, this dissertation presents new computational approaches at high resolution to problems related to structure-based drug discovery with unprecedented accuracy. Given such high accuracy, these approaches are very promising in addressing practical issues in pharmaceutical research and development, and in enhancing our capability in the search for new safe drugs.