Biophysical Society Thematic Meeting | Hamburg 2022

Biophysics at the Dawn of Exascale Computers

Poster Abstracts

39-POS Board 39 GPU-AIDED OPTIMIZATION OF HIGH-FIDELITY LIPID FORCE FIELDS Markus S. Miettinen 1,2 ; Hanne Antila 1 ; 1 Max Planck Institute of Colloids and Interfaces, Theory & Bio-Systems, Potsdam, Germany 2 University of Bergen, Chemistry, Bergen, Norway The NMRlipids community (nmrlipids.blogspot.fi)—an open science project devoted to benchmarking current lipid force field against nuclear magnetic resonance (NMR) data— has showed that none of the current lipid molecular dynamics (MD) models (force fields) captures the correct conformational ensemble of lipids in bilayers [1,2]. More recently, we used the open databank of bilayer simulations predominantly originating from the NMRlipids project, to show that the conformational dynamics of the models are not fully correct either [3], although the best models capture the experimental data well. The correct representation of both the conformations and the dynamics is a crucial prerequisites when drawing conclusions from the simulations. Unfortunately, further development of the MD models is hindered by the overwhelming workload, lack of comprehensive comparison to experiments, and outdated approaches (such as hand-tuning parameters). We address these issues by combining atomistic resolution data (NMR C-H order parameters) with an evolutionary optimization algorithm to create a semi-automated force field building tool. As each optimization round requires running MD simulations to test the force field candidates against the experimental target, the approach is solely made feasible by the usage of GPU-accelerated MD engine. Using the algorithm, we have created the first POPC (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine) models that reproduce the conformational ensemble as described by the C–H order parameters and have dynamics comparable to the best existing force fields (CHARMM36 and Slipids). Additionally, we have validated the models against other types of experimental data and assessed their response to changing conditions.[1] Botan et al., J. Phys. Chem. B, 199(49), 2015.[2] Antila et al., J. Phys. Chem. B, 123(43), 2019.[3] Antila et al., J. Chem. Inf. Model, 61(2), 2021.

100