Data-Driven Many-Body Potentials for Chemically Accurate Simulations from Molecules to Biomolecules

F. Paesani

Achieving chemical accuracy in molecular simulations of aqueous systems has remained a central challenge for decades due to the interplay of many-body interactions and nuclear quantum effects. In this contribution, I will introduce a unified, data-driven many-body (MB) formalism that has enabled realistic simulations from small gas-phase clusters to bulk solutions and interfaces. Our data-driven many-body energy (MB-nrg) potentials, such as MB-pol and its extensions, systematically integrate physics-based representations with machine-learned components trained on coupled-cluster reference data. The MB-nrg approach not only reproduces structural, thermodynamic, and spectroscopic properties across phases, but also predicts emergent phenomena such as the location of the liquid–liquid critical point in supercooled water. Building on this foundation, I will present recent applications that extend this framework from water and aqueous solutions to generic covalently bonded molecules, demonstrating predictive biomolecular simulations with chemical accuracy