TY - JOUR
T1 - Accurate Estimation of Protein Folding and Unfolding Times
T2 - Beyond Markov State Models
AU - Suárez, Ernesto
AU - Adelman, Joshua L.
AU - Zuckerman, Daniel M.
N1 - Publisher Copyright:
© 2016 American Chemical Society.
PY - 2016/8/9
Y1 - 2016/8/9
N2 - Because standard molecular dynamics (MD) simulations are unable to access time scales of interest in complex biomolecular systems, it is common to "stitch together" information from multiple shorter trajectories using approximate Markov state model (MSM) analysis. However, MSMs may require significant tuning and can yield biased results. Here, by analyzing some of the longest protein MD data sets available (>100 μs per protein), we show that estimators constructed based on exact non-Markovian (NM) principles can yield significantly improved mean first-passage times (MFPTs) for protein folding and unfolding. In some cases, MSM bias of more than an order of magnitude can be corrected when identical trajectory data are reanalyzed by non-Markovian approaches. The NM analysis includes "history" information, higher order time correlations compared to MSMs, that is available in every MD trajectory. The NM strategy is insensitive to fine details of the states used and works well when a fine time-discretization (i.e., small "lag time") is used.
AB - Because standard molecular dynamics (MD) simulations are unable to access time scales of interest in complex biomolecular systems, it is common to "stitch together" information from multiple shorter trajectories using approximate Markov state model (MSM) analysis. However, MSMs may require significant tuning and can yield biased results. Here, by analyzing some of the longest protein MD data sets available (>100 μs per protein), we show that estimators constructed based on exact non-Markovian (NM) principles can yield significantly improved mean first-passage times (MFPTs) for protein folding and unfolding. In some cases, MSM bias of more than an order of magnitude can be corrected when identical trajectory data are reanalyzed by non-Markovian approaches. The NM analysis includes "history" information, higher order time correlations compared to MSMs, that is available in every MD trajectory. The NM strategy is insensitive to fine details of the states used and works well when a fine time-discretization (i.e., small "lag time") is used.
UR - http://www.scopus.com/inward/record.url?scp=84981541494&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84981541494&partnerID=8YFLogxK
U2 - 10.1021/acs.jctc.6b00339
DO - 10.1021/acs.jctc.6b00339
M3 - Article
C2 - 27340835
AN - SCOPUS:84981541494
SN - 1549-9618
VL - 12
SP - 3473
EP - 3481
JO - Journal of Chemical Theory and Computation
JF - Journal of Chemical Theory and Computation
IS - 8
ER -