Self-Learning Adaptive Umbrella Sampling Method for the Determination of Free Energy Landscapes in Multiple Dimensions

The potential of mean force describing conformational changes of biomolecules is a central quantity for understanding the function of biomolecular systems. Calculating an energy landscape of a process that depends on three or more reaction coordinates requires extensive computational power, making some multidimensional calculations practically impossible. Here, we present an efficient automatized umbrella sampling strategy for calculating a multidimensional potential of mean force. The method progressively learns by itself, through a feedback mechanism, which regions of a multidimensional space are worth exploring and automatically generates a set of umbrella sampling windows that is adapted to the system. The selflearning adaptive umbrella sampling method is first explained with illustrative examples based on simplified reduced model systems and then applied to two nontrivial situations: the conformational equilibrium of the pentapeptide Met-enkephalin in solution and ion permeation in the KcsA potassium channel. With this method, it is demonstrated that a significant smaller number of umbrella windows needs to be employed to characterize the free energy landscape over the most relevant regions without any loss in accuracy.