Performance of Universal Machine-Learned Potentials with Explicit Long-Range Interactions in Biomolecular Simulations

Universal machine-learned potentials promise transferable accuracy across compositional and vibrational degrees of freedom, yet their application to biomolecular simulations remains underexplored. This work systematically evaluates equivariant message-passing architectures trained on the SPICE-v2 data set with and without explicit long-range dispersion and electrostatics. We assess the impact of model size, training data composition, and electrostatic treatment across in- and out-of-distribution benchmark data sets, as well as molecular simulations of bulk liquid water, aqueous NaCl solutions, and biomolecules, including alanine tripeptide, the mini-protein Trp-cage, and Crambin. While larger models improve accuracy on benchmark data sets, this trend does not consistently extend to properties obtained from simulations. Predicted properties also depend on the composition of the training data set. Long-range electrostatics show no systematic impact across systems. However, for Trp-cage, their inclusion yields increased conformational variability. Our results suggest that imbalanced data sets and immature evaluation practices currently challenge the applicability of universal machine-learned potentials to biomolecular simulations.