Accelerating Replica Exchange Molecular Dynamics: A Comparison of Hydrogen Mass Repartitioning and Light Water Models

TL;DR

This paper compares two methods for speeding up molecular simulations and finds that one method, hydrogen mass repartitioning, is more efficient than the other.

Contribution

The study introduces a hybrid light water model and compares computational efficiency of hydrogen mass repartitioning and light water models in replica exchange simulations.

Findings

HMR2 and HMR3 models are up to 4-fold more computationally efficient than the HMR1 reference.

The hLW model improves computational efficiency and better reproduces energetic and conformational properties than LW.

HMR3 is preferable to LW due to broader applicability and simplicity.

Abstract

Accelerating conformational sampling through changes
in molecular
mass is an attractive option in biomolecular modeling. Here, we examine
the utility and compare the efficiency of hydrogen mass repartitioning
(HMR) and light water (LW) models in the context of replica exchange
(RE) simulations of an alanine dipeptide. To maintain integrator stability,
we introduced scaling of integration steps with RE temperatures and
determined their maximum values, assuring the stability of RE simulations.
HMR2 and HMR3 models featuring doubled and tripled hydrogen masses
and, to a lesser extent, the LW model reproduce the energetic and
conformational properties of alanine dipeptide in water compared to
the HMR1 reference. This conclusion is based on comparing kinetic
and potential energies, free energy landscapes of the peptide, as
well as its structural properties, including hydrogen bonding, water
counts in the peptide first solvation shell, and RMSD distributions.
Thereby, our results demonstrate that both HMR and LW models can be
integrated into RE simulations. We then compared HMR and LW models
with respect to the computational efforts required to equilibrate
alanine dipeptide. HMR2 and HMR3 are up to 4-fold more computationally
efficient than the HMR1 reference, whereas LW lags behind being less
than a factor of 2 more efficient. As a result, LW efficiency is 2-fold
lower than that of HMR3. This outcome means that increasing the integration
step provides faster sampling than boosting water diffusion. Even
if the computation of long-range interactions is adjusted with the
length of the integration step and the step in LW simulations is further
increased, the model remains less efficient than HMR3. We considered
a hybrid variant of LW, hLW, featuring heavier water and mass repartitioning
applied to all hydrogens, affording longer integration steps than
LW does. hLW improves computational efficiency and provides more accurate
reproduction of energetic and conformational properties of alanine
dipeptide than LW. We concluded that HMR3 and hLW models demonstrate
good performance in replica exchange simulation, but the former is
preferable due to broader applicability and simplicity. hLW remains
an excellent alternative to HMR3, but its scope is limited to “water-rich”
systems. More generally, our findings suggest that among the two approaches,
HMR or decreasing water mass, the former is more effective. Since
LW simulations are not currently supported out-of-the-box by the NAMD
molecular dynamics program, we implemented a patch enabling LW functionality.

Figures