Multiple branched adaptive steered molecular dynamics

Steered molecular dynamics, SMD, [S. Park and K. Schulten, J. Chem. Phys. 120, 5946 (2004)] combined with Jarzynski's equality has been used widely in generating free energy profiles for various biological problems, e.g., protein folding and ligand binding. However, the calculated averages are generally dominated by "rare events" from the ensemble of nonequilibrium trajectories. The recently proposed adaptive steered molecular dynamics, ASMD, introduced a new idea for selecting important events and eliminating the non-contributing trajectories, thus decreasing the overall computation needed. ASMD was shown to reduce the number of trajectories needed by a factor of 10 in a benchmarking study of decaalanine stretching. Here we propose a novel, highly efficient "multiple branching" (MB) version, MB-ASMD, which obtains a more complete enhanced sampling of the important trajectories, while still eliminating non-contributing segments. Compared to selecting a single configuration in ASMD, MB-ASMD offers to select multiple configurations at each segment along the reaction coordinate based on the distribution of work trajectories. We show that MB-ASMD has all benefits of ASMD such as faster convergence of the PMF even when pulling 1000 times faster than the reversible limit while greatly reducing the probability of getting trapped in a non-significant path. We also analyze the hydrogen bond breaking within the decaalanine peptide as we force the helix into a random coil and confirm ASMD results with less noise in the numerical averages.