Adaptive Sampling of Markov-state models with MD Simulations to assess the rates of biologically relevant processes

It is a challenge for classical molecular dynamics to directly simulate the time and length scales of biologically relevant processes. Incomplete sampling makes the calculation of macroscopic observables such as rates for major state transitions difficult to estimate correctly. A number of techniques bias the energy landscape to accelerate processes that are difficult to reach with direct simulation. However the altered landscape introduces artifacts that obscure rate estimates. Here we show how adaptive-sampling workflows that iteratively redirect swarms of trajectories can enhance the sampling of the slowest kinetic processes. Markov-state models are then adaptively refined with continual updates from these parallel swarms resulting in unbiased, well-sampled estimates of kinetic rates. We demonstrate how this method can be applied to provide long-timescale kinetics for atomistic simulations of protein-protein interactions.