Tool-Augmented AI for Drug Discovery: An Agentic Platform for Integrated Molecule Generation and Screening

Over the past few decades, numerous scientific tools have been developed for computer-aided drug discovery (CADD). These tools typically require domain experts for their proper use and interpretation of results. The goal of this project is to create an agentic planner for CADD so that the user can start with natural language prompts which will automatically trigger tools to generate, screen or explain drugs and drug-like molecules. In details, a master workflow has been designed by domain experts to guide agent planners. An in-house question-and-answer (Q&A) dataset was generated to supplement public datasets to ensure higher accuracy of agent deployments. Multiple new tools have been developed to connect with mature capabilities in the field. We show that our agentic solution can assist scientists in initiating CADD projects, finding hits/leads, performing lead optimization with no prior knowledge of programming skills and/or with a significant improvement of efficiency.