Learning Drug Specifity in Protein Families by Docking

DESCRIPTION (provided by applicant): The goal is to provide a set of powerful docking-based rational drug discovery tools to take advantage of the increasingly rich amount of information on protein families sequence and structure emerging from genomics and structural genomics efforts. The tools will improve the reliability of current virtual screening methods by taking advantage of all available information on a protein target's family, including aligned sequences, available structures, co-crystalized ligands, and active compounds described in the literature. The tools will also specifically allow common regions of family active sites to be targeted in the design of family-focused combinatorial libraries with high activity rates against any member of the family. Lastly the tools will allow unique regions of the target active site to guide the design of compounds with high selectivity for a specific target. A key component of this suite of tools is a novel description of the interaction of a ligand with the surface of a protein called a footprint. Footprints are used as input to clustering, filtering, and learning methods to analyze the results of docking screens and to compare docking results across members of the target family. Virtual screening of large libraries of chemical compounds will be performed on four protein families, the data analyzed, and new promising scaffolds for future focused combinatorial libraries will be detected. In summary the successful development and application of the tools described in this grant request can: 1. Significantly increase the success rate of virtual screens; 2. Generate highly effective focused combinatorial libraries to protein families; 3. Generate target-specific leads in difficult target families such as kinases; 4. Guide the construction of highly focused in-vitro experimental screening.