Calculations of solvation free energies and pK<sub>a</sub>s in diverse solvents provide an immense value for understanding general chemical and physical processes, but generalized approaches that accurately account for explicit solvent interactions are often difficult to parameterize and/or are computationally demanding. There is is also interest in using machine learning (ML) forcefields that would drive molecular simulations with quantum chemical accuracy at a significantly reduce cost. In this presentation, we will report our latest results extending the gradient-domain machine learning (mbGDML) method, a kernel method that uses a global descriptor, using a many-body formalism to enable size transferable simulations of solvents. Our three-body potentials accurately reproduce quantum chemical forces and energies with far fewer training data than what is needed in alternative SchNet, GAP, and PIP approaches, and results on bulk solvents are promisingly consistent with reference data. We will also discuss how the many-body implementation also brings several compelling benefits including the straightforward enabling of thermodynamic integration over alchemical reaction pathways and show progress toward applications of predicting solvation free energies of ions. Benchmarking results compared to other available methods and the promise for mbGDML in future applications will be discussed.

Rough and ready solvation simulations using many-body gradient domain machine learning

Where thousands of chemistry professionals meet to share ideas and advance scientific and technical knowledge. Spring 2023 abstracts, posters, and presentations will launch at the end of March.

ACS Spring 2023

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.