(168g) Using Multistate Reweighting to Rapidly Explore Molecular Parameter Space

We use a molecular benchmark set and the multistate Bennett acceptance ratio (MBAR) reweighting formalism to explore the potential energy parameter space of non-bonded interactions. The choice of simulation parameters such as the method of treating long range electrostatics and van der Waals terms can affect the accuracy of thermodynamic quantities calculated in molecular dynamics and Monte Carlo simulations. We show how we can reweight a small number of simulations using multistate reweighting techniques to rapidly predict free energies and enthalpies with thousands of different parameter combinations rather than run individual simulations for each parameter combination. We use these reweighting techniques to quickly identify the parameter set giving the least deviation in observables from expensive parameter choices for a given amount of simulation time.   

We also show how these reweighting methods can rapidly prototype large numbers of possible potential energies choices for molecular systems. The key to the accuracy and efficiency of these multistate reweighting techniques is that instead of reweighting from a single sampled state, we simultaneously reweight from a number of sampled states in a way that maximally uses information from all simulations.