Automating bespoke torsion parameterization on Orion ®

The accuracy of force fields is vital in several molecular modeling efforts, including studying protein-ligand interactions for drug discovery. Despite having far fewer parameters than traditional force fields, those released by the Open Force Field (OpenFF) Initiative have been shown to accurately reproduce geometries and energies obtained from quantum chemistry (QC) calculations. However, the sensitivity of torsion parameters to local chemical environments in molecules sometimes leads to the underperformance of force fields for molecules with novel chemistry. Bespoke fitting of torsion parameters has been discussed as a solution to this problem. It has been shown that bespoke fitting of torsion parameters can improve the accuracy of property prediction, such as binding free energies.

In this work, we have developed a protocol for bespoke fitting of the torsion parameters for small molecules based on QC calculations. The QC torsion scans are performed on fragments constructed around rotatable bonds that are minimal yet chemically representative, thus reducing the computational cost of generating the reference data. Our effort of bespoke fitting torsion parameters of OpenFF force fields differs from the existing solutions in that we use a hybrid HF-3C/DFT QC protocol and a newly developed efficient parameter optimization procedure. We apply our newly fitted parameters to perform binding free energy calculations of small molecules using Non-Equilibrium switching (NES). We show that bespoke fitting of torsion parameters improves the accuracy of binding free energy predictions obtained using NES for multiple datasets with known binding affinities. Applying bespoke fitting of torsion parameters to improve binding free energy predictions via NES poses a promising pathway for more successful drug discovery campaigns.