PFP/MM: A Hybrid Approach Combining a Universal Neural Network Potential with Classical Force Fields for Large-Scale Reactive Simulations

Universal machine-learning interatomic potentials (uMLIPs) enable reactive molecular simulations with near-DFT accuracy, yet applying them efficiently to large, realistic condensed-phase systems remains computationally demanding. Here we present PFP/MM, a hybrid approach that combines a uMLIP, PreFerred Potential (PFP), with molecular mechanics (MM) to enable both large-scale and long-time simulations that are challenging for uMLIP-only calculations. Using an alanine dipeptide in explicit water, we achieve multi-ns/day enhanced sampling and obtain a Ramachandran plot consistent with established basins. For an intramolecular nucleophilic addition reaction in a polar solvent environment, we reproduce the expected solvent-induced stabilization in the free-energy profile. We further apply the approach to a cytochrome P450 Compound I hydroxylation reaction and obtain a free-energy landscape consistent with the accepted reaction mechanism. These results demonstrate that uMLIP-based reactive simulations can be applied to diverse condensed-phase processes in large, realistic environments.