Efficiency of adaptive temperature-based replica exchange for sampling large-scale protein conformational transitions

Abstract

Temperature‐based replica exchange (RE) is now considered a principal technique for enhanced sampling of protein conformations. It is also recognized that existence of sharp cooperative transitions (such as protein folding/unfolding) can lead to temperature exchange bottlenecks and significantly reduce the sampling efficiency. Here, we revisit two adaptive temperature‐based RE protocols, namely, exchange equalization (EE) and current maximization (CM), that were previously examined using atomistic simulations (Lee and Olson, J. Chem. Physics, 134, 24111 (2011)). Both protocols aim to overcome exchange bottlenecks by adaptively adjusting the simulation temperatures, either to achieve uniform exchange rates (in EE) or to maximize temperature diffusion (CM). By designing a realistic yet computationally tractable coarse‐grained protein model, one can sample many reversible folding/unfolding transitions using conventional constant temperature molecular dynamics (MD), standard REMD, EE‐REMD, and CM‐REMD. This allows rigorous evaluation of the sampling efficiency, by directly comparing the rates of folding/unfolding transitions and convergence of various thermodynamic properties of interest. The results demonstrate that both EE and CM can indeed enhance temperature diffusion compared to standard RE, by ~3‐ and over 10‐fold, respectively. Surprisingly, the rates of reversible folding/unfolding transitions are similar in all three RE protocols. The convergence rates of several key thermodynamic properties, including the folding stability and various 1D and 2D free energy surfaces, are also similar. Therefore, the efficiency of RE protocols does not appear to be limited by temperature diffusion, but by the inherent rates of spontaneous large‐scale conformational re‐arrangements. This is particularly true considering that virtually all RE simulations of proteins in practice involve exchange attempt frequencies (~psˉ¹) that are several orders of magnitude faster than the slowest protein motions (~μsˉ¹). Our results also suggest that the efficiency of RE will not likely be improved by other protocols that aim to accelerate exchange or temperature diffusion. Instead, protocols with some types of guided tempering will likely be necessary to drive faster large‐scale conformational transitions.