| The immense amount of data generated since the onset of the post-genomic era has affected all fields of computational biology, including the study of protein-protein interactions. Databases of experimentally identified protein complexes provide a gold standard test for developing accurate models of undiscovered protein complexes. However, protein-protein docking methods still suffer from the prevalence of false positives among their results. Using evolutionary conservation information and artificial intelligence techniques, this thesis proposes four related methods for obtaining more native-like conformations in protein-protein docking as well as detecting residues that are critical for structure and function of the protein. First, two stochastic methods for refining docked dimeric and multimeric complexes are introduced. Then, a novel machine learning based tool is presented to predict the structural similarity of a docked protein complex to its native form. Using this tool for ranking decoys, a third method is proposed to refine docked protein complexes. Finally, combining evolutionary conservation information with protein rigidity analysis, another machine-learning based method is presented for predicting critical protein residues, which can play an important role in protein-protein binding. |
