Speaker
Description
Molecular Dynamics simulations are a valuable tool for investigating a wide variety of phenomena at the atomistic scale. However, it is this very high resolution that limits the time scales accessible to such simulations. Typically, this limitation is overcome by deploying methods that require knowledge of the system's metastable states and an approximation of the reaction coordinate, referred to as a collective variable.
Here, we propose a new enhanced sampling method which requires neither.
We build our approach on the observation that in rarely occurring events like phase transition or protein folding processes, there is a trade-off between enthalpy and entropy. It is rare fluctuations in these quantities which eventually result in a transition. Thus, we describe the state of the system via the distributions of single particle energies or of appropriately defined per-atom entropies. We then devise an algorithm that favours sampling the distribution tails, thus facilitating high-energy barrier crossing and the discovery of new states, with null or minimal prior knowledge of the system.
We demonstrate the generality of our approach by successfully simulating protein folding, ligand binding to receptors, and crystallising an undercooled metal melt starting just from the unfolded state, the apo state, and the liquid state respectively.