A Direct Method for Incorporating Experimental Data into Multiscale Coarse-Grained Models

To extract meaningful data from molecular simulations, it is necessary to incorporate new experimental observations as they become available. Recently, a new method was developed for incorporating experimental observations into molecular simulations, called experiment directed simulation (EDS), which utilizes a maximum entropy argument to bias an existing model to agree with experimental observations while changing the original model by a minimal amount. However, there is no discussion in the literature of whether or not the minimal bias systematically and generally improves the model by creating agreement with the experiment. In this work, we show that the relative entropy of the biased system with respect to an ideal target is always reduced by the application of a minimal bias, such as the one utilized by EDS. Using all-atom simulations that have been biased with EDS, one can then easily and rapidly improve a bottom-up multiscale coarse-grained (MS-CG) model without the need for a time-consuming reparametrization of the underlying atomistic force field. Furthermore, the improvement given by the many-body interactions introduced by the EDS bias can be maintained after being projected down to effective two-body MS-CG interactions. The result of this analysis is a new paradigm in coarse-grained modeling and simulation in which the "bottom-up" and "top-down" approaches are combined within a single, rigorous formalism based on statistical mechanics. The utility of building the resulting EDS-MS-CG models is demonstrated on two molecular systems: liquid methanol and ethylene carbonate.