Bottom-up coarse-grained (CG) molecular dynamics models are parameterized using complex effective Hamiltonians. These models are typically optimized to approximate high dimensional data from atomistic simulations. However, 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 classification can be used to variationally estimate high dimensional error and that explainable machine learning can help convey this information to scientists. This approach is demonstrated using Shapley additive explanations and two CG protein models. This framework may also be valuable for ascertaining whether allosteric effects at the atomistic level are accurately propagated to a CG model.
Using classifiers to understand coarse-grained models and their fidelity with the underlying all-atom systems
Current address: Department of Mathematics and Computer Science, Freie Universität Berlin, Arnimallee 12, 14195 Berlin, Germany
Note: This paper is part of the JCP Special Topic on Machine Learning Hits Molecular Simulations.
Aleksander E. P. Durumeric, Gregory A. Voth; Using classifiers to understand coarse-grained models and their fidelity with the underlying all-atom systems. J. Chem. Phys. 21 June 2023; 158 (23): 234101. https://doi.org/10.1063/5.0146812
Download citation file: