Machine Learning-Driven Approaches to Identifying Slow Dynamics of Biomolecular Systems

Machine Learning-Driven Approaches to Identifying Slow Dynamics of Biomolecular Systems

Myongin Oh - Dr. Oh is a candidate for the Assistant Professor in Computational Protein Biochemistry position

Monday January 24, 2022

Time: 1-2 PM

Direct link Webex:

https://mun.webex.com/mun/j.php?MTID=md9b3075191f9f310ddac0655ca237c4a

Abstract

Machine Learning-Driven Approaches to Identifying Slow Dynamics of Biomolecular Systems

Biomolecules are not rigid and static but flexible and dynamic. Biomolecular dynamics has been extensively investigated using an array of experimental and computational techniques to identify slow motions central to their functions. Classical molecular dynamics simulation is a powerful tool for the study of biomolecular systems to understand their behaviour by integrating Newton’s equations of motion at the molecular scale. A central problem in existing simulations is that biomolecules exhibit complex free energy landscapes in which long-lived metastable states separated by large energy barriers are present. The kinetic bottlenecks impede transitions between metastable states and restrict the timescale that can be explored, although slow motions are generally considered to play an important role in biomolecular functions. To resolve this sampling issue, researchers have incorporated machine learning into biomolecular simulations. In this talk, I will first highlight recent advances in machine learning-driven approaches to identifying slow dynamics of biomolecular systems. I will then discuss how to combine time-lagged independent component analysis, an unsupervised dimensionality reduction algorithm, with metadynamics, an enhanced sampling technique, to find optimal collective variables for molecular simulations of membrane permeation of a small drug molecule, trimethoprim, as an example.