Abstract

Many physical, chemical or biological processes of interest involve studying transitions of a system between metastable states separated by high free energy barriers. Due to these barriers, such transitions rarely occur in conventional molecular dynamics simulations. A plethora of enhanced sampling techniques has been developed in the past decade to mitigate this problem. These techniques, however, were designed either to augment the number of transitions or to calculate dynamic properties. Recently, we developed a new enhanced sampling method that allows one to calculate static and dynamic properties. In our method, Gaussian Mixture Based Enhanced Sampling, the known metastable states are discriminated against using a set of descriptors. The fluctuations of these descriptors in the different states are then fitted to Gaussian Mixtures. Then, a bias potential is built from this fit such that it can enhance the fluctuations of the system only within the states while allowing the system to transition between states in an unbiased manner. We exemplified the effectiveness of our method in cases of a chemical reaction and ligand unbinding.