Computational prediction and analysis of macromolecular interactions

Abstract

Protein interactions regulate gene expression, cell signaling, catalysis, and many other functions across all of molecular biology. We must understand them quantitatively, and experimental methods have provided the data that form the basis of our current understanding. They remain our most accurate tools. However, their low efficiency and high cost leave room for predictive, computational approaches that can provide faster and more detailed answers to biological problems. A rigid-body simulation can quickly and effectively calculate the predicted interaction energy between two molecular structures in proximity. The fast Fourier-transform-based mapping algorithm FTMap predicts small molecule binding 'hot spots' on a protein's surface and can provide likely orientations of specific ligands of interest that may occupy those hot spots. This process now allows unique ligands to be used by this algorithm while permitting additional small molecular cofactors to remain in their bound conformation. By keeping the cofactors bound, FTMap can reduce false positives where the algorithm identifies a true, but incorrect, ligand pocket where the known cofactor already binds. A related algorithm, ClusPro, can evaluate interaction energies for billions of docked conformations of macromolecular structures. The work reported in this thesis can predict protein-polysaccharide interactions and the software now contains a publicly available feature for predicting protein-heparin interactions. In addition, a new approach for determining regions of predicted activity on a protein's surface allows prediction of a protein-protein interface. This new tool can also identify the interface in encounter complexes formed by the process of protein association—more closely resembling the biological nature of the interaction than the former, calculated, binary, bound and unbound states.