Replica Exchange Molecular dynamics (REMD) is a widely used sampling method to obtain converged ensembles of biomolecules. While REMD can provide converged ensembles, however, the convergence rate of REMD is largely dependent on the speed of sampling a Boltzmann-weighted ensemble at high temperature, which is still a difficult problem for large biomolecules. To fix this problem, we previously developed a variant of REMD, namely non-Boltzmann Reservoir REMD (RREMD), in which an unweighted (non-Boltzmann) set of pre-generated structure snapshots representing different local minima are used as a reservoir. Monte Carlo moves are then attempted between the reservoir structures and the replicas. This method resulted in much faster ensemble convergence rates at all temperatures compared to conventional REMD.<br/>In this work, we ported the RREMD code onto GPUs making it 20x faster than the CPU code. This facilitated us in carrying out extensive simulations to test the method on a wide variety of systems. We show that with careful selection of structure snapshots corresponding to different local minima, the method can reproduce accurate Boltzmann ensembles obtained by much more expensive conventional REMD, with at least 25x faster convergence rates even for larger proteins (&gt;50 amino acids) compared to conventional REMD. Furthermore, with the new GPU code, we obtained converged ensembles for &gt;50 amino acid proteins within just 10 hours (wall clock time) on a GTX 980.<br/>Finally, since it is not necessary for the reservoir to be Boltzmann-weighted, structures obtained from non-MD methods such as homology-based models or Rosetta-based models can be integrated into this method, making it a powerful tool that combines physics-based approaches with non-physics-based approaches leading to better structure predictions of biomolecules. The GPU RREMD code is now available for use with the AMBER suite of programs.

Improving REMD efficiency with Monte Carlo moves using a structure reservoir

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.