Adaptive steered molecular dynamics of biomolecules

ABSTRACT

Adaptive steered molecular dynamics (ASMD) is a variant of steered molecular dynamics (SMD) in which the driving of the auxiliary – viz steered – particle is performed in stages. In SMD, many nonequilibrium trajectories are generated allowing for fast sampling of the ensemble space and access to rare events. Equilibrium observables, such as the potential of mean force along the pathway, result from averaging over these trajectories using Jarzynski's Equality (JE). Unfortunately, in SMD, a large number of trajectories are needed to cover all possible configurations in order to obtain converged quantities in the exponential average of the JE, and this is computationally expensive. ASMD reduces the number of trajectories that must be sampled by discarding those trajectories that have deviated far from the equilibrium path in stages. At the end of a stage, one chooses – or contracts – one (in naïve ASMD) or some (in multi-branched ASMD or MB-ASMD) of the configurations produced from the previous stage to initiate the trajectories in the next stage. Alternatively, in full-relaxation ASMD (FR-ASMD), all generated structures are relaxed under the constraint of a fixed auxiliary particle exerting no net work on the system. We provide a direct comparison of the energetics and other observables obtained from these approaches. We find that FR-ASMD is preferred when the unfolding pathways follow up along a single funnel and the system is sufficiently small that computational resources are not a limiting concern. It gives the highest accuracy in such cases while avoiding the inefficiencies of SMD. However, for complex energy landscapes typical of most multi-domain proteins, MB-ASMD is preferred because it provides a mechanism to sample alternative pathways while suffering only a modest loss in accuracy compared to FR-ASMD.

KEYWORDS:

• Adaptive steered molecular dynamics
• Jarzynski's equality
• peptide unfolding

Acknowledgments

This work has been partially supported by the National Science Foundation (NSF) through Grant No. CHE 1700749. The computing resources necessary for this research were provided in part by the National Science Foundation through XSEDE resources under Grant Number CTS090079, and by the Maryland Advanced Research Computing Center (MARCC).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work has been partially supported by the National Science Foundation (NSF) [grant number CHE 1700749]. The computing resources necessary for this research were provided in part by the National Science Foundation through XSEDE resources [grant number CTS090079], and by the Maryland Advanced Research Computing Center (MARCC).