Developing and Applying Accessible Computational Technologies for the Design of Small-Molecule Ligands

Project Summary/Abstract Drug discovery is tremendously costly and failure-prone, complicating advances that would otherwise im- prove human health. Computer-aided drug discovery (CADD) accelerates this process by predicting molecules with pharmacological potential before synthesis and testing. These bioactive molecules also serve as chemical probes for studying fundamental biological processes. But despite notable successes, CADD methods still suf- fer in terms of accuracy and accessibility. The Durrant lab addresses these challenges by developing innovative, user-friendly technologies to (1) identify new bioactive small molecules and (2) improve their phar- macological properties, with the ultimate goal of helping the scientific community develop new therapies for patients in need. In recent years, we have developed tools that characterize the surface pockets of disease-implicated pro- teins where such compounds (e.g., drugs) interact. These tools identify pocket locations and predict pocket flexibility. We have also developed modeling techniques to better predict protein/molecule binding geometries (binding poses), as well as methods that suggest chemical additions to improve binding (lead optimization). These methods broadly apply to many proteins, so our research is not tied to a single system or disease. Still, to rigorously validate our techniques, we have partnered with experimentalists studying diverse proteins impli- cated in protein synthesis, metabolism, cancer, and viral infections. Our work has been cited thousands of times, demonstrating impact. Further, our online tools have been accessed over 100,000 times, serving as a useful community resource. The lab has attracted funding from university, government, and industry partners. In the next five years, we will continue to develop innovative CADD methods while expanding in new direc- tions. Our focus is on improving the accuracy, accessibility, and broad adoption of CADD approaches. First, we will create a big-data database of predicted protein-ligand complexes to improve the accuracy of structure- based machine-learning models. This database will greatly expand the structural data available for training and so will serve as a helpful community resource. Second, we will use the database to develop new methods for (1) assessing how potently a small-molecule ligand interacts with its binding pocket and (2) suggesting lead- optimization strategies tailored to specific proteins. Third, we will create an online system for deploying novel CADD tools through the web browser, enabling streamlined workflows and enhancing usability. This online system will further broaden impact by (1) making it easier to translate our browser-based tools into other lan- guages, (2) ensuring the tools are accessible for those with disabilities, and (3) providing new mentoring opportunities for students. Finally, through collaborations with partners in academia and industry, we will also continue to validate our tools by applying them to biologically interesting test proteins.