Molecular: Parallel sampling of functional motions in proteins

Molecular dynamics can reveal molecular motion in atomistic detail, but this comes at a high computational cost. Billions of time steps of femtoseconds are required to reach biologically relevant time scales of microseconds. For particular scientific problems smarter sampling could accelerate sampling by orders of magnitude, thereby enabling new problems to be tackled. Understanding functional motions is the key to understanding how many biomolecules function. Such transitions are often slow because of by high free-energy barriers, which means straightforward simulations will not capture the transitions. In case one has some idea of the direction(s) of functional motions, one can try to modify the sampling along such directions.

We have adapted and extended a weight histogram method, developed originally developed for particle physics applications by Jack Lidmar at KTH. This method will sample a flat distribution along one more target degrees of freedom, thereby overcoming the issue of the exponential decreasing probability of crossing barriers during normal Boltzmann sampling. This can result in an acceleration of the sampling by one or more orders of magnitude, depending on the height of the barriers and the friction in the system. Additionally, with this method multiple replicas can be simulated simultaneously, where the sampling is continuously accumulated over all replicas. The allows for very efficient parallelization (on top of the usual parallelization for each replica), but even more importantly for shorter times to solution.