Title:	Establishment of best practices for FMM-based MD simulations of electrosprayed protein complexes and application to ferritin
DNr:	NAISS 2023/5-416
Project Type:	NAISS Medium Compute
Principal Investigator:	Erik Marklund <erik.marklund@kemi.uu.se>
Affiliation:	Uppsala universitet
Duration:	2023-11-01 – 2024-05-01
Classification:	10603 10402
Keywords:

Abstract

MD simulations of gas-phase proteins have become increasingly common over the past decade, following the breakthrough of native mass spectrometry (MS), which allows for intact protein complexes to be separated and analysed in the gas-phase. Recent advances in using native MS to select subpopulations of proteins from a large ensemble and deposit them on electron microscopy grids will likely increase the demand for such simulations further. The gas-phase conditions are very challenging for conventional MD methods however, and many commonly used methods in the condensed phase are ill-suited for vacuum conditions. The currently best approach leads to O(N^2) scaling, which severely limits the time and length scales that can be investigated with gas-phase MD. Recently, an implementation of the fast multipole method (FMM) was made in the Gromacs simulation package, which could be a very good alternative for our applications. The algorithm scales as O(N), and is well suited for systems with highly inhomogeneous particle density, like proteins in the gas phase. We have successfully made FMM-Gromacs run on Tetralith and done a number of performance benchmarks to see which settings work best for different types of systems. FMM-Gromacs outperforms other approaches for most system sizes, especially for the largest complexes, for which the high performance is the most valuable. We need to finish our benchmarks, including our less developed accuracy tests, in order to establish and publish best practices for using FMM-Gromacs for electrosprayed and otherwise aerosolised protein complexes. We will also apply FMM-Gromacs to two ferritin systems for which different collaborators have interesting data the we hope to explain with our simulations.