Neural potentials of proteins extrapolate beyond training data

We evaluate neural network (NN) coarse-grained (CG) force fields compared to traditional CG molecular mechanics force fields. We conclude NN force fields are able to extrapolate and sample from unseen regions of the free energy surface when trained with limited data. Our results come from 88 NN force fields trained on different combinations of clustered free energy surfaces from 4 protein mapped trajectories. We used a statistical measure named total variation similarity (TVS) to assess the agreement between free energy surfaces of mapped atomistic simulations and CG simulations from the trained NN force fields. Our conclusions support the hypothesis that constructing force fields on one region of the protein free energy surface can indeed extrapolate to unexplored regions. Additionally, the force matching error was found to only be weakly correlated with a force field's ability to reconstruct the correct free energy surface.