Accelerating molecular dynamics simulations of lipid bilayers

Dnr:

SNIC 2016/1-407

Type:

SNAC Medium

Principal Investigator:

Per Larsson

Affiliation:

Uppsala universitet

Start Date:

2016-09-23

End Date:

2017-10-01

Primary Classification:

10203: Bioinformatik (beräkningsbiologi) (tillämpningar under 10610)

Webpage:

Allocation

Abstract

Molecular dynamics simulations of pure lipid bilayers, and of more complex systems and processes with embedded proteins, membrane fusion and active and passive membrane transport are ubiquitous these days. The ever-increasing power of the computers used to carry out these systems means that the accessible spatial and temporal scales that can be studied also continue to increase. For classical simulations with all-atom force fields, the fastest motions are bond vibrations involving hydrogen atoms, followed by bond vibrations involving only heavy atoms. To get around this in simulations, constraints are often used and considered a more faithful representation of the physical behavior of bond vibrations which are almost exclusively in their vibrational ground state. Simply constraining all angles involving hydrogens is not recommended: Apart from the large computational cost due to highly coupled constraints, constraining all angles involving hydrogens indirectly constrains the angles between heavy atoms, which leads to too rigid molecules as well as slow dynamics. A more elegant solution instead replaces the hydrogen atoms with massless virtual interaction sites, the positions of which are a function of the coordinates of the heavy atoms. In this project, we want to implement and investigate the applicability of such virtual interaction sites for a number of commonly used all-atoms forcefield. While such techniques have been extensively used for simulations of proteins, relatively little attention has been paid to applying it for lipids. This is important since it can potentially increase the accessible time-scales for simulations of bilayers by a factor of almost 3, at no extra computational cost.