Efficiency of Replica Exchange Simulations in Determining Thermodynamic Properties of Biological Systems
Terry J. Schmitt and Thomas A. Knotts IV. Department of Chemical Engineering, Brigham Young University, 350 CB, Provo, UT 84602
In recent years, the replica exchange simulation technique has become commonplace when calculating thermodynamic properties, such as protein folding free energies, of complex biological systems. Recently, however, the efficiency of the technique compared to standard molecular dynamics has been called into question. Previous studies have sought answers in this regard by measuring the amount of sampling produced by each technique as a function of total simulation time. However, questions remain because 1) the metrics by which sampling is defined do not necessarily lead to accurate thermodynamic properties and 2) simulation time does not exactly correspond to CPU time due to the computational overhead associated with swapping. To remove these uncertainties, we present work on how efficient replica exchange is at calculating the thermodynamic properties of the system (rather than sampling efficiency) as a function of CPU time (rather than simulation time). We discuss both the precision and accuracy of the two methods in determining the potential energy, heat capacity, and free energy of folding of both coarse-grained and atomistic proteins. We also address how the swapping frequency affects performance. Taken as a whole, the results demonstrate that, in general, replica exchange produces accurate and precise values of thermodynamic properties more efficiently than molecular dynamics, but that the degree of improvement is specific to each property.