Pocket optimization and its application to identify small-molecule inhibitors of protein-protein interactions

Because of their ubiquitous nature in many cellular processes, modulating protein-protein interactions offers tremendous therapeutic potential. However, protein-protein interactions remain a difficult class of drug targets, as most protein interaction sites have not evolved to bind small molecules. Indeed, some protein interaction sites are thought to be simply not amenable to binding any small molecule at all. Other sites feature small molecule binding pockets that simply are not present in the unbound or protein-bound conformations, making structure-based drug discovery difficult. Sometimes, inhibitors bind to multiple family members with high affinity, causing toxicity. In this dissertation I seek to address many of these challenges, by developing methodologies to assess the druggability of a target, assess the selectivity of known inhibitors, identify conformations that are sampled uniquely by a single protein, and identify inhibitors of protein-protein interactions. To assess druggability, I developed the “pocket optimization” protocol which uses a biasing potential to create an ensemble of conformations that contain pockets at a specified location on the protein surface. I showed that low-resolution, low energy inhibitor shapes are encoded at druggable sites and sampled through low-energy fluctuations, whereas they are not present at random sites on protein surfaces. To assess selectivity and screen for inhibitors, I developed “exemplars”, representations of a pocket based on the perfect “non-physical” complementary ligand, allowing the comparison of pocket shapes independent of protein sequence. I predicted the selectivity of an array of inhibitors to a related family of proteins by comparing the exemplars from the known small-molecule bound conformation to the ensemble of exemplars from a “pocket optimized” ensemble. I identified distinct conformations that could be targeted for identifying selective inhibitors de novo by comparing ensembles of exemplars from related family members to one another. Finally, I developed a screening protocol that uses the speed of exemplar versus small molecule comparisons to screen very large compound libraries against ensembles of distinct, “pocket optimized” pocket conformations