ANC Seminar - Andrea Rizzi

Tuesday, 7th March 2023

Accelerating atomistic simulations with machine learning

Abstract: Molecular simulations are a popular tool in biology and drug discovery for their ability to predict biophysical quantities (e.g., drug-protein binding affinities and kinetics). To this end, the main challenge is often to sample exhaustively from a complex multimodal distribution (i.e., the Boltzmann distribution) which describes the probability of finding the molecular system in a given conformation. Indeed, while this distribution is known from quantum mechanical laws, these are in practice too expensive to be simulated, and the distribution must necessarily be modelled by introducing approximations and tradeoffs between accuracy and complexity/computational cost. In this talk, I will describe how ML has impacted molecular simulations by helping tame this complexity, focusing in particular on two examples. I will show how normalizing flow neural networks can be used to quickly recover the accuracy of quantum mechanical models starting from simulations using cheap and qualitative models. Finally, I will briefly describe how ML can be used to identify slow degrees of freedom and escape local minima through enhanced sampling techniques.

Bio: Andrea Rizzi is a postdoctoral researcher at the Institute for Computational Biomedicine in Forschungszentrum Juelich and the Atomistic Simulation group at the Italian Institute of Technology. He obtained his PhD in 2020 from the Tri-Institutional PhD. Program in Computational Biology and Medicine (Cornell University, Weill Cornell University, and Memorial Sloan-Kettering Cancer Center) working on physics-based approaches for the prediction of ligand-receptor binding affinities. Prior to his PhD, he worked on automatic vehicle collision avoidance as an undergraduate researcher at the Massachusetts Institute of Technology, and he obtained a Master of Science from Politecnico di Milan. His research focuses on developing machine learning techniques for biomolecular simulations with a particular focus on drug discovery applications.

Event type: Seminar

Date: Tuesday, 7th March 2023

Time: 11:00

Location: G.03

Speaker(s): Andrea Rizzi 

Chair/Host: Douglas Armstrong