Coarse-graining and efficiently sampling with autoencoders.

3^rd February 2023, 14:00
Gabriel Stoltz
CERMICS, Ecole des Ponts ParisTech, France

Abstract

A coarse-grained description of atomistic systems in molecular dynamics is provided by reaction coordinates. These nonlinear functions of the atomic positions are a basic ingredient to compute more efficiently average properties of the system of interest, such as free energy profiles. However, reaction coordinates are often based on an intuitive understanding of the system, and one would like to complement this intuition or even replace it with automated tools. One appealing tool is autoencoders, for which the bottleneck layer provides a low dimensional representation of high dimensional atomistic systems. In order to have an efficient numerical method, autoencoders should be combined with importance sampling techniques based on adaptive biasing methods. The algorithm then iterates between an update of the reaction coordinate, and free energy biasing. I will discuss some mathematical foundations of this method, and present illustrative applications for biophysical systems, including alanine dipeptide and chignolin. Some on-going extensions to more demanding systems, namely HSP90, will also be hinted at.