Matthew Harrigan and Robert McGibbon in the Pande Group have recently made a movie which highlights how MSM Adaptive sampling works in a simple yet relevant example (a 2 dimensional Muller potential). Adaptive sampling works by first running several relatively short trajectories, then building an MSM model, then using the MSM to run new trajectories.
In this movie, we compare the adaptive sampling approach to the more traditional approach of a single, long trajectory, and find that long trajectories can get stuck in minima and the adaptive sampling approach can quickly explore the whole space.
Here’s what’s going on. First, we start from some point. In this movie, this is the energy minima in the lower right, but this could be the unfolded state of the protein or the inactive state of the kinase. We then run many, relatively short trajectories from that spot. Next, we build an MSM. This is shown by a “Voronoi Tessellation” in the movie, dividing the space explored so far into MSM states. Then, using the MSM, new starting configurations for trajectories are chosen (large colored dots) and new trajectories are run from those new points. This process continues until convergence.