July 12

Nick Charron (presenting on behalf of Cecilia Clementi, Freie Universität Berlin, Germany)

14:55 - 15:30

Molecular simulation has traditionally been utilized as a "computational microscope" for complex systems - leading to important advances in chemistry, medicine, and biology through key atomic-level insights and analysis. However, for many systems, molecular simulation is unable to efficiently sample timescales associated with biologically relevant events, such as protein folding, aggregation, ligand binding, or complex assembly. Even with rapid advances in computer hardware, the speed of traditional molecular dynamics is limited by fundamental numerical and physical constraints. As an answer to this problem, we present recent efforts in developing and applying coarse grain molecular models that improve simulation efficiency while maintaining physical consistencies with the original all-atom systems using powerful machine learning techniques. Through a variety of such approaches, we show how physics-guided machine learning can result in robust and powerful predictive molecular models.