The torsion energy term describes the energy of a molecule as it rotates around its bonds and is commonly given as a sum of cosine terms with periodicities that are chosen using chemical expertise. To decrease the computational cost of fitting torsions in large molecules, it is common practice to fragment them into smaller entities that should reproduce the chemistry about each torsion of interest. However, most existing cheminformatics fragmentation schemes fragment molecules for synthetic accessibility and do not consider electronic properties. In order to automate and generalize the process, several problems need to be addressed. These are: intelligent fragmentation of compounds for quantum chemistry (QC) calculations to reduce misrepresentation of the chemical environment of the torsion; automation of many QC torsion scans; automated parameter fitting and model selection, finding the correct granularity of chemical environments for transferable torsion types. This talk focuses on the open-source and modular software developed to solve these problems within Open Force Field Initiative framework. The QC calculations are captured, indexed with several canonical chemical identifiers, and deposited to the open QCArchive database for future use.

This work was supported by MolSSI Seed and Investment Fellowship awarded to Chaya Stern (https://molssi.org/the-molssi-software-fellowship-program/).