Year: 2026
Type: preprint
Abstract: The choice of explicit water model may affect absolute binding free energy (ABFE) predictions for bromodomains, whose conserved acetyl-lysine pockets contain structured waters. We assembled a benchmark of ten distinct bromodomain–ligand complexes with experimental isothermal titration calorimetry thermodynamics and compared four TIP-family water models, TIP3P, TIP4P, TIP5P, and TIP7P, using a common alchemical protocol. TIP3P gave the lowest point-estimate mean unsigned error (MUE), 1.56 kcal mol−1, against 1.89–2.11 kcal mol−1 for the other models, at the lowest cost. TIP7P correlated best with experiment (r = 0.73), but the experimental affinities span only 2.1 kcal mol−1, about the size of the errors of the models. The 0.55 kcal mol−1 MUE difference between TIP3P and TIP7P had a nominal 95% paired bootstrap interval that excluded zero when only the benchmark composition was resampled, but not when the calculated affinities were perturbed by their sampling-sensitivity measures or the four charged ligands were left out. No other paired difference was resolved, including after finite-size corrections for the charged ligands. The leg decomposition shows that solvent and complex contributions can compensate, so water-model effects must be judged over the complete binding cycle. With one calculation per complex and model, the higher-site TIP models gave no demonstrated accuracy benefit with this force field for 1.4–2.7 times the cost.
Source: ChemRxiv
Institution: Ege University
Language: English
Subfield: Drug Discovery
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7223180932