Expanded ensemble predictions of solvent–water partition coefficients for drug-like molecules using OpenFF 2.0 and GAFF 2.1
Abstract
How well do general-purpose molecular mechanics force fields predict solvent–water partition coefficients for drug-like molecules? To answer this question, we performed massively parallel expanded ensemble free energy calculations using distributed computing to evaluate two widely used force fields—GAFF 2.1 and OpenFF 2.0—across a diverse set of 70 molecules and solvents (chloroform, dodecane, octanol and toluene). We find both force fields have comparable accuracy, with predicted transfer free energies having mean unsigned errors (MUEs) below 2 kcal/mol in nearly all cases. Our results also reveal systematic error, likely due to the known deficiencies of fixed-charge force fields. We expect that this dataset and our approach to systematic evaluation using distributed computing may be useful toward future efforts in force field validation and parameterization.
Related
- Extrapolating Collisional Rate Coefficients for Molecules of Astrophysical Interest Using Neural Networks
- Determination of Hydroquinone Clathrate Stability for Different Guest Molecules Using NPH Ensemble Molecular Dynamics.
- Carbon dioxide expanded ionic liquid mixtures for changing reaction outcomes and efficient separation
- Antiaromatic expanded porphyrin enables redox-tunable photoconductivity in columnar liquid crystals
- Expanded use of benzamidines in the Groebke–Blackburn–Bienaymé synthesis of imidazoles and their biological evaluation
BibTeX
@misc{goold2026expanded,
title = {{Expanded ensemble predictions of solvent–water partition coefficients for drug-like molecules using OpenFF 2.0 and GAFF 2.1}},
author = {Steven R. Goold and Vincent A. Voelz},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010089/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010089/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Expanded ensemble predictions of solvent–water partition coefficients for drug-like molecules using OpenFF 2.0 and GAFF 2.1
AU - Steven R. Goold
AU - Vincent A. Voelz
PY - 2026
DA - 2026/10/08
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010089/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010089/v1
AB - How well do general-purpose molecular mechanics force fields predict solvent–water partition coefficients for drug-like molecules? To answer this question, we performed massively parallel expanded ensemble free energy calculations using distributed computing to evaluate two widely used force fields—GAFF 2.1 and OpenFF 2.0—across a diverse set of 70 molecules and solvents (chloroform, dodecane, octanol and toluene). We find both force fields have comparable accuracy, with predicted transfer free energies having mean unsigned errors (MUEs) below 2 kcal/mol in nearly all cases. Our results also reveal systematic error, likely due to the known deficiencies of fixed-charge force fields. We expect that this dataset and our approach to systematic evaluation using distributed computing may be useful toward future efforts in force field validation and parameterization.
ER -