Ensuring Mixing Efficiency of Replica-Exchange Molecular Dynamics Simulations.

Abstract

We address the question of constructing a protocol for replica-exchange molecular dynamics (REMD) simulations that make efficient use of the replica space, assess whether published applications are achieving such "mixing" efficiency, and provide a how-to guide to assist users to plan efficient REMD simulations. To address our first question, we introduce and discuss three metrics for assessing the number of replica-exchange attempts required to justify the use of a replica scheme and define a "transit number" as the lower bound for the length of an efficient simulation. Our literature survey of applications of REMD simulations of peptides in explicit solvent indicated that authors are not routinely reporting sufficient details of their simulation protocols to allow readers to make independent assessments of the impact of the method on their results, particularly whether mixing efficiency has been achieved. Necessary details include the expected or observed replica-exchange probability, together with the total number of exchange attempts, the exchange period, and estimates of the autocorrelation time of the potential energy. Our analysis of cases where the necessary information was reported suggests that in many of these simulations there are insufficient exchanges attempted or an insufficiently long period between them to provide confidence that the simulation length justifies the size of the replica scheme. We suggest guidelines for designing REMD simulation protocols to ensure mixing efficiency. Two key recommendations are that the exchange period should in general be larger than 1 ps and the number of exchange attempts should be chosen to significantly exceed the transit number for the replica scheme.