New preprint: environment-aware Lennard-Jones parameters for ML/MM

Our preprint Atomic Environment Aware Lennard-Jones Parameterization for Electrostatic Embedding ML/MM Simulations is now available on ChemRxiv.

Electrostatics in ML/MM simulations have been studied extensively, but exchange-repulsion and dispersion are still usually described with Lennard-Jones parameters taken from standard force fields, which are not consistent with the electrostatic embedding model. We derive Lennard-Jones parameters from the exchange-hole dipole moment (XDM) model and the Fedorov polarizability scaling relation, and propose a one-shot procedure to fine-tune them to experimental free energies. For the electrostatic machine learning embedding (EMLE) scheme, the resulting models match the accuracy of conventional force fields and mechanical embedding, while giving parameters that adapt to the chemical environment of each atom. This opens a route to more accurate reactive simulations, where dispersion coefficients can change strongly along the reaction.

Work by João Morado, Kirill Zinovjev, Lester O. Hedges, Julien Michel, and Daniel J. Cole, in collaboration with the University of Edinburgh, OpenBioSim and Newcastle University.