TCBG Seminar

AI-driven multi-scale simulations for COVID-19 drug discovery

Dr. Arvind Ramanathan
Data Science and Learning Division
Argonne National Lab
Lemont, IL

Monday, January 25, 2021
3:00 pm
Zoom webinar link

Abstract

Artificial intelligence (AI) and machine learning (ML) techniques have been widely successful in a variety of structural biology applications, including protein structure prediction and acceleration of multiscale molecular simulations. In this talk, we highlight how we have been applying AI/ML methods to discover inhibitors against the novel coronavirus disease (COVID-19) causing agent, namely the severe acute respiratory coronavirus 2 (SARS-CoV-2). In particular, we describe DeepDriveMD, which leverages deep learning algorithms to learn latent representations from long time-scale MD simulations, and adaptively samples conformations that lead to more productive trajectories at scale. We show that using DeepDriveMD, with only 12% of time spent in sampling, we could identify over 25% more (novel) conformations of the Spike protein and its interactions with the ACE2 receptor. We also highlight the use AI/ML in discovering novel inhibitors for the SARS- CoV-2 MPro, where our initial virtual screening, followed by AI-driven simulations provided insights into subsequent rounds of design, leading to improved affinity to the target protein. We also discuss some potential pitfalls and challenges that we faced as part of our implementation and conclude with our perspective on how AI/ML methods could be truly beneficial for driving multiscale simulations with emerging supercomputing platforms.


More information wwww.anl.gov/profile/arvind-ramanathan-0


Main TCBG Seminars page