A GPU-Accelerated Fast Multipole Method for Gromacs. Performance and Accuracy

Bartosz Kohnke,Carsten Kutzner,Helmut Grubmuller

doi:10.1016/j.bpj.2020.11.1242

Bartosz Kohnke, Carsten Kutzner + Show 1 more

Open Access

https://doi.org/10.1016/j.bpj.2020.11.1242

Copy DOI

Abstract

The evaluation of Coulomb forces is the computationally most expensive task in molecular dynamics (MD) simulations. Most simulation codes rely on the Particle Mesh Ewald (PME) method to evaluate these forces. This is for a good reason - PME is extremely fast. However, due to the underlying Fourier transformations, PME suffers from a communication bottleneck which prohibits efficient scaling to large numbers of compute nodes. The fast multipole method (FMM) is an alternative to PME that by construction needs significantly less communication volume in parallel.

Highlights

859-Pos Single Particle Tracking Model Segmentation using Change Detection Techniques Boris I
We found a large number of potential antibody escape mutations in the receptor binding domain (RBD), some of which agree with other studies
We found a smaller number of potential antibody strengthening mutations in Ab that could be used to improve the therapeutic value of Ab

Summary

Introduction

859-Pos Single Particle Tracking Model Segmentation using Change Detection Techniques Boris I. Molecular dynamics simulations were used to generate trajectory snapshots. These snapshots were used as inputs for FoldX, a fast semi-empirical method for estimating folding and binding free energies. We found a large number of potential antibody escape mutations in the RBD (i.e., those predicted to destabilize RBD-Ab interactions), some of which agree with other studies.

Results

Conclusion