Multiscale Neural Network Potential for the Simulation of Protein Dynamics and Enzymatic Reactions

Zoom password: ai4science

The on-site event will be followed by fika in the Analysen coffee area (fika from 16:00-16:30).

Abstract:

From simple clustering techniques to sophisticated neural networks, the use of machine learning has become a valuable tool in many fields of chemistry in the past decades. Here, we describe how we explore the use of machine learning (ML) for predict physical interactions between particles in molecular dynamics (MD) simulations in order to simulate biological systems in the high nanosecond range at QM level of accuracy.

We have developed a graph neural network approach using anisotropic message passing (AMP) that is specifically suited for the use in ML/MM simulations (analogous to QM/MM) with electrostatic embedding. While the first versions were system-specific, the AMP approach showed already local transferability among similar systems, which we exploited in the recent third version where we trained a foundational model on 1.5 million QM/MM reference data points of peptides, small molecules, and transition states.

Using this model, we could perform hundred nanosecond simulations of proteins in water and compute free-energy profiles of enzymatic reactions where the “QM zone” included the full protein.

Speaker:

Sereina Riniker, ETH Zurich.