Starts 2 Mar 2022 16:30
Ends 2 Mar 2022 17:30
Central European Time
via Zoom
Carter Tribley Butts
(University of California, Irvine, USA)
Advances in computational chemistry have allowed us to simulate molecular systems of ever-increasing complexity over ever longer time scales, but making sense of the resulting trajectory data remains a challenge.  Here, I describe a number of ways that kernel learning can be used as a tool for gaining insights from molecular dynamics simulations, with a particular eye to identifying features of use for subsequent computational or experimental investigation.  Specific applications include segmentation of macromolecular conformation spaces, identifying and characterizing separatrices separating trajectories bound for distinct product states, and characterizing the primary axes of variation in protein active site conformations.  In conjunction with the latter, I also describe how kernels can be used to capture feature spaces of particular theoretical interest, as illustrated by an application to the analysis of the active site dynamics of the SARS-CoV-2 main protease.