📝 SummarXiv

Abstract

Accurate absolute binding free energy (ABFE) calculations can reduce the time and cost of identifying drug candidates from a diverse pool of molecules that may have been overlooked experimentally. These calculations typically employ explicit solvents; however these models can be computationally demanding and challenging to implement due to the difficulties in sampling waters and managing changes in net charge. To address these challenges, we introduce an automated parallel Python workflow that adopts the double decoupling method, incorporating conformational restraints and pairing it with the implicit generalized Born (GB) solvation model. This approach enhances convergence, reduces computational costs and avoids the technical issues associated with explicit solvents. We applied this workflow to a series of 93 host-guest complexes from the TapRoom database. When pooling all systems, the GB(OBC) model correlated well with experiment ($R^2 = 0.86$); however, this global metric obscured much weaker correlations observed within individual hosts ($R^2 = 0.3-0.8$). Systematic errors associated with charged functional groups (notably ammonium and carboxylates) were also evident, resulting in root-mean-squared errors (RMSEs) greater than 6.12 kcal/mol across all models. Although the GB models did not achieve reliable accuracy across all systems, they may be practical when all ligands contain the same functional groups. Furthermore, a linear correction based on these functional groups reduced RMSE values to within 1 kcal/mol of experiment, and our error analysis suggests specific changes for future GB models to improve accuracy. Overall, the Python workflow demonstrates promise for fast, reliable absolute binding free energy calculations by leveraging the sampling efficiency of GB in a fully automated application.