Computing Transitions in Macromolecular Systems: Dynamic Importance Sampling

Abstract

Understanding and predicting conformational change in macromolecules is central to linking structure and function. Performing straight-forward all-atom molecular dynamics would, in principle, enable sampling of conformational changes. However, the time-scale for functionally important transitions, exceeds the usual molecular dynamics timescales by several orders of magnitude. For example, with large amounts of computer time all these transitions could be observed with good statistics and the results collected simply by waiting long enough. Thus to sample on longer time-scales requires the development of biased molecular dynamics methods, where the bias can be applied and corrected for at the end. In our approach, called ‘Dynamic Importance Sampling’ we generate a series of independent trajectories that are conditioned on starting and ending in defined conformations. Trajectories are generated using two different algorithms: one uses a soft-racheting scheme based on stochastic trajectories and the other uses information from the set of normal modes. The algorithms, which require no initial pathway, are capable of rapidly determining multiple pathways between known states. The associated probablity scores, determined by correcting for the bias, allows us to rank order the most likely pathways. We will present examples from three-helix bundles and other systems for both analysis and possible experimental work.