Abstract: We first provide a high-level overview of different directions in modeling molecular interactions where deep learning methods have already had significant positive real-world impact. Then we focus on the particular task of predicting the structure in which a small molecule binds to a protein as a case study by investigating how we address it with our DiffDock method. DiffDock is a diffusion generative model for predicting binding structures — a fundamental task in structural biology — which we will discuss in greater detail.

Bio:  Hannes Stärk is a first-year PhD student at MIT in the CS and AI Laboratory (CSAIL). He works on geometric deep learning and physics-inspired ML and applications in molecular biology and other physical systems.