Development And Application Of Pseudoreceptor Methods

Quantitative Structure-Activity Relationship (QSAR) methods are a commonly used tool in the drug discovery process. These methods attempt to form a statistical model that relates descriptor properties of a ligand to the activity of that ligand compound towards a specific desired physiological response. QSAR methods are known as a ligand-based method, as they specifically use information from ligands and not protein structural data. However, a derivation of QSAR methods are pseudoreceptor methods. Pseudoreceptor methods go beyond standard QSAR by building a model representation of the protein pocket. However, the ability of pseudoreceptors to accurately replicate natural protein surfaces has not been studied. The goal of this thesis work is to investigate the necessary descriptors to map a protein binding pocket and a method to accurately recreate the 3-D spatial structure of the binding pocket. In addition, additional applications of existing pseudoreceptor methods are explored. To identify the necessary descriptors to map a protein binding pocket, we developed a program that decomposes the protein-ligand interaction surface from a large number of ligand-bound protein crystal structures. The binding pockets of the protein structure are identified, and then the physico-chemical properties of the protein are mapped onto the solvent accessible surface of the binding pocket. A number of 2-D Gaussian functions are then placed onto this surface to model the protein’s physico-chemical properties. We found that a small number of these Gaussians were able to accurately replicate the properties of the protein. With this knowledge, we then desired a means of accurately recreating the binding pocket surfaces of proteins only the structures of their bound ligands. Typically in pseudoreceptor methods either the average or combined solvent accessible surface of the ligand set is used. To test this, we generated iso-level surfaces of the solvent accessible surfaces of sets of ligands for which the co-crystallized protein structure is available. We also tested additional sets of surfaces located beyond the ligand’s solvent accessible surface. We found that any single surface was unable to accurately reproduce the protein-ligand interaction surface, and multi-surface approach using numerous iso-surfaces is needed to accurately represent the protein. Finally, we explored the application of RAPTOR, an existing pseudoreceptor method, to the problem of the prediction of Sites-of-Metabolism (SoM) for Cytochrome P450s (Cyps). In our approach, we used RAPTOR as a means of discriminating between active (correctly predicted SoM) docking poses of ligands from decoy (incorrect SoM) poses. With our method, we achieved the highest reported rate of SoM prediction across nine Cyp isoforms, with the best reported performance on seven of those nine isoforms