Wei Zhang - Convergence of replica exchange molecular dynamics.

Version 1

      Publication Details (including relevant citation   information):

      J Chem Phys. 2005 Oct   15;123(15):154105.


            Replica exchange molecular   dynamics (REMD) method is one of the generalized-ensemble   algorithms which performs random walk in energy space and helps a   system to escape from local energy traps. In this work, we   studied the accuracy and efficiency of REMD by examining its   ability to reproduce the results of multiple extended   conventional molecular dynamics (MD) simulations and to enhance   conformational sampling. Two sets of REMD simulations with   different initial configurations, one from the fully extended and   the other from fully helical conformations, were conducted on a   fast-folding 21-amino-acid peptide with a continuum solvent   model. Remarkably, the two REMD simulation sets started to   converge even within 1.0 ns, despite their dramatically different   starting conformations. In contrast, the conventional MD within   the same time and with identical starting conformations did not   show obvious signs of convergence. Excellent convergence between   the REMD sets for T>300 K was observed after 14.0 ns REMD   simulations as measured by the average helicity and free-energy   profiles. We also conducted a set of 45 MD simulations at nine   different temperatures with each trajectory simulated to 100.0   and 200.0 ns. An excellent agreement between the REMD and the   extended MD simulation results was observed for T>300 K,   showing that REMD can accurately reproduce long-time MD results   with high efficiency. The autocorrelation times of the calculated   helicity demonstrate that REMD can significantly enhance the   sampling efficiency by 14.3+/-6.4, 35.1+/-0.2, and 71.5+/-20.4   times at, respectively, approximately 360, approximately 300, and   approximately 275 K in comparison to the regular MD. Convergence   was less satisfactory at low temperatures (T<300 K) and a slow   oscillatory behavior suggests that longer simulation time was   needed to reach equilibrium. Other technical issues, including   choice of exchange frequency, were also examined.

      Address (URL): http://www.ncbi.nlm.nih.gov/pubmed/16252940