Evaluating the Functional Importance of Conformer-Dependent Atomic Partial Charge Assignment

Abstract

Physics-based methods such as protein-ligand binding free energy calculations have been increasingly adopted in early-stage drug discovery to prioritize promising compounds for synthesis. However, the accuracy of these methods is highly dependent on the details of the calculation and choices made while preparing the ligands and protein ahead of running calculations. During ligand preparation, researchers typically assign partial atomic charges to each ligand atom using a specific ligand conformation for charge assignment, often the input conformer. While it is a well-known problem that partial charge assignment is dependent on conformation, little investigation has explored the downstream effects of varied partial charge assignment on free energy estimates. Preliminary benchmarks from the Open Free Energy Project show that generating partial charges from different input conformers leads to variation of up to $\pm 5.3$ kcal/mol in calculated relative binding free energies due to variation in partial charges alone. In this study, we more systematically explore this issue, investigating it in smaller systems using absolute hydration free energy calculations to reduce the degrees of freedom and sources of statistical error as compared to larger protein-ligand systems. We investigate how differences in partial charge generation (such as those caused by input conformer choice, partial charge engine, and hardware) may lead to differences in calculated absolute hydration free energy (AHFE) values. We demonstrate that supplying different input conformers to a partial charge engine can result in atomic partial charge discrepancies of up to 0.681 e, resulting in differences in calculated AHFE of $6.9\pm 0.1$ kcal/mol. We find that even relatively small variations in partial charge assignment can result in notable differences in calculated AHFE. Thus, care should be taken when assigning partial charges to ensure reproducibility and accuracy of any resulting free energy calculations. We expect that these effects will be magnified in pharmaceutically relevant binding free energy calculations with additional degrees of freedom, more highly directional interactions than in water, and potentially more statistical error.