学术文献库

Synthetic molecular dynamics (synMD) trajectories from learned generative models have been proposed as a useful addition to the biomolecular simulation toolbox. The computational expense of explicitly integrating the equations of motion in molecular dynamics currently is a severe limit on the number and length of trajectories which can be generated for complex systems. Approximate, but more computationally efficient, generative models can be used in place of explicit integration of the equations of motion, and can produce meaningful trajectories at greatly reduced computational cost. Here, we demonstrate a very simple synMD approach using a fine-grained Markov state model (MSM) with states mapped to specific atomistic configurations, which provides an exactly solvable reference. We anticipate this simple approach will enable rapid, effective testing of enhanced sampling algorithms in highly non-trivial models for both equilibrium and non-equilibrium problems. We demonstrate the use of a MSM to generate atomistic synMD trajectories for the fast-folding miniprotein Trp-cage, at a rate of over 200 milliseconds per day on a standard workstation. We employ a non-standard clustering for MSM generation that appears to better preserve kinetic properties at shorter lag times than a conventional MSM. We also show a parallelizable workflow that backmaps discrete synMD trajectories to full-coordinate representations at dynamic resolution for efficient analysis.