Status: Active

Drug discovery remains one of the most resource-intensive endeavors in biomedicine, with estimates placing the average cost of bringing a single new therapeutic to market at over $2 billion and spanning more than a decade. A key rate-limiting step in this pipeline is the identification of small molecules that can effectively bind to a target protein of interest—a process that must navigate the immense and sparsely populated space of drug-like chemical structures. Traditional experimental screening approaches, while powerful, are constrained by cost, time, and physical library size, making them impractical for exhaustively exploring chemical space. Virtual screening offers a scalable solution, functioning as a funnel that narrows billions of candidate molecules through successive layers of increasing computational fidelity. At the broad end of the funnel, we’ve developed fast, high-throughput ML-based screening tools—such as ConPLex and CoNCISE—that prioritize candidates based on predicted binding potential. These models allow us to efficiently reduce the search space from billions of existing compounds to tens of thousands of potentially viable compounds in a matter of seconds. On the bottom end of the funnel, we apply more computationally expensive but precise docking simulations, which are typically limited to a few thousand molecules due to resource constraints. To bridge the two ends pf the funnels, we endeavor to develop an easy-to-use pipeline, in which we develop and apply various filtering algorithms to selectively re-rank and prioritize the top ~1,000 molecules for molecular docking. Following docking, we select the most promising candidates for experimental validation, achieving a dramatic reduction in cost while retaining screening accuracy.

Research supported by: Duke Whitehead Scholar Award

Principal Investigator: Rohit Singh