(311b) The Open Force Field “Rosemary” Release and New Approaches to Chemically Transferable Force Fields

The Open Force Field Initiative (OpenFF), at https://openforcefield.org, is academic/industrial collaboration 1) developing extensible, open source toolkits for constructing, applying, and evaluating force fields; 2) curating public datasets necessary to build high-accuracy biomolecular force fields; and 3) generating improved molecular force fields for biomolecular and other soft matter applications. The effort is designed to improve the reproducibility and openness of the development of classical force fields for quantitative molecular exploration, prediction, and design.

In this presentation, I will discuss the Open Force Field version 3.0 force field, “Rosemary,” the first Open Force Field version to include support for proteins. In particular, Rosemary parameters, similar to previous OpenFF force fields, are transferable; the same parameters are identically applied to chemical moieties in small molecules, biopolymers and other chemical environments of interest.

I will discuss the force field developments that make it possible to support polymers in OpenFF including graph convolutional neural network approaches to assign charges that scales linearly with molecular size, chemical matching of monomers within larger macromolecules, surrogate modeling approaches to speed and even improve the optimization of nonbonded force field parameters, and graph convolutional neural net approaches to define continuous chemical environments.

I will also discuss our broader collaborative effort to develop validation sets to determine the extent to which a force field accurately models protein structure, protein-ligand binding, and small molecule crystal thermodynamics.