Development of Computational Methods for Predicting Protein Interactions

Abstract

Proteins are important components of living organisms and are involved in many biological processes. The biological functions of proteins result from their molecular interactions with other molecules such as metal ions, small organic compounds, peptides, lipids, nucleic acids, or other proteins. Therefore, computational approaches to predict interactions between proteins and other molecules are useful to understand protein functions in molecular level and to design molecules that regulate protein functions. Specifically, ligand binding site prediction methods can be used to identify druggable sites of target proteins while protein-ligand docking techniques can contribute to identifying hit or lead compounds and optimizing lead compounds during structure-based drug discovery process. In addition, because a large fraction of cellular proteins self-assemble to form symmetric homo-oligomers to play their biological roles, computational methods to predict homo-oligomer structures can also contribute to drug discovery process by providing atomic details of target oligomer interfaces.<br/> In this thesis, three computational methods developed to predict protein interactions are introduced: (1) an improved metal and organic molecule binding site prediction method, (2) a protein-ligand docking method with an improved hybrid scoring function and a sampling algorithm utilizing predicted binding hot spot information, and (3) a protein homo-oligomer modeling method using bioinformatics and physical chemistry approaches. All methods described here show high performances in benchmark tests when compared to other state-of-the-art programs. These benchmark results suggest that computational approaches introduced in this thesis can be applied to in silico drug discovery process.