BPS2026 – Development and benchmarking of the open force field initiative self-consistent force field for proteins and small molecules

Proteins containing non-canonical residues play critical roles in cellular biology and drug design. Developing force field (FF) parameters for these proteins is challenging and time-consuming; often no straightforward, automated process can extend an existing FF to model new chemistries. Additionally, parameters introduced to model biomolecules are seldom benchmarked against data used to train parameters from general FFs, leaving any inconsistencies between general and biomolecule-specific parameters unassessed. There is a need for an extensible framework to derive and benchmark FF parameters for diverse chemistries in proteins and small molecules self-consistently. We describe the development and benchmarking of a self-consistent FF for proteins and small molecules from the open force field initiative (OpenFF). OpenFF assigns parameters using direct chemical perception via the SMIRKS cheminformatics language, providing extensive coverage of chemical space with fewer parameters than conventional atom-typed FFs. We explore to what extent protein-specific parameters not shared with small molecules are needed. Starting from general OpenFF parameters, we trained several parameter sets containing different numbers of protein-specific parameters against quantum mechanical (QM) data for both small molecules and short peptides. While QM-trained FFs reproduce validation QM and NMR data sets for short peptides, they understabilize helices in folded proteins. We then tuned protein-specific backbone torsions by reweighting against NMR observables for the folded protein GB3 and the helical peptide (AAQAA)3. Finally, we validated our parameters against small molecule QM and physical properties, additional protein NMR experiments, and protein-ligand binding free energies. Our self-consistent FF can model both small molecule ligands and proteins, including those with non-canonical residues, using a single set of parameters. We anticipate that this FF will facilitate research and drug development that requires modeling non-canonical proteins to understand their biological or pharmacological roles.