Explaining classifiers to understand coarse-grained models

Abstract

Bottom-up coarse-grained molecular dynamics models are parameterized using complex effective Hamiltonians. These models are typically optimized to approximate high dimensional data from atomistic simulations. In contrast, human validation of these models is often limited to low dimensional statistics that do not necessarily differentiate between the CG model and said atomistic simulations. We propose that explainable machine learning can directly convey high-dimensional error to scientists and use Shapley additive explanations do so in two coarse-grained protein models.