Machine learning for modelling solvent effects | Department of Chemistry

The Duarte group at Oxford Chemistry has introduced a new approach for incorporating solvents into the modelling of chemical reaction by machine learning potentials (MLP), recently published in Nature Communications.

The proposed strategy ensures that the training sets are generated efficiently and have reduced levels of redundancy.

Their strategy yielded accurate and stable MLPs for a test Diels-Alder reaction modelled in explicit water and methanol. These MLPs were used to propagate molecular dynamics simulation that would take several years using standard ab initio methods.

The resulting MLPs provide key insights into the origin of solvation effects on the Diels-Alder reaction and pave the way to a broader exploration of solvent effects on chemical reactivity.

You can read more in the group’s Nature blog post about the study, or in the original paper that was recently published in Nature Communications.

Graphical abstract for "Modelling chemical processes in explicit solvents with machine learning potentials", showing a Diels-Alder reaction modelled in water and methanol solvents.