3DRISM Solvation for Accurate and Large-Scale Protein Hydration Prediction and Mechanistic Interpretation

Solvation plays essential roles in protein stability, molecular recognition, and enzyme catalysis. The three-dimensional reference interaction site model (3DRISM), an integral equation theory of molecular solvation, provides an attractive balance between computational efficiency and predictive accuracy by explicitly predicting 3-dimensional hydration structures while avoiding the costly sampling required by explicit-solvent molecular dynamics (MD) simulations. However, its performance for protein-wide hydration prediction has not been systematically evaluated across a large and diverse set of protein structures, particularly in comparison with recent machine learning (ML) approaches. Here, we systematically benchmark 3DRISM against the state-of-the-art ML model, SuperWater, across its Train and Test datasets and an independently constructed Query dataset of 3,139 protein structures outside its training domain. Across these datasets, 3DRISM achieves hydration-site prediction accuracy comparable to SuperWater. 3DRISM calculations performed with our EPISOL software package remain computationally efficient for large-scale applications. Comparison with explicit-solvent MD simulations of 20 randomly selected proteins further demonstrates that 3DRISM accurately reproduces equilibrium solvent density distributions across diverse local protein environments. Finally, using the periplasmic binding protein SiaP as a case study, we show that the solvent density predicted by 3DRISM directly captures mutationinduced solvent depletion within a buried ligand-binding pocket, and accurately reconstructs the corresponding water positions involved in the hydrogen-bond network. By directly computing the equilibrium solvent distribution, 3DRISM provides mechanistic insight into ligand recognition by revealing hydration changes that are difficult to capture using conventional explicit-solvent MD simulations potentially due to the slow water exchange in the buried binding pocket. Our results establish 3DRISM as an accurate, computationally efficient, and mechanistically informative framework for large-scale protein hydration prediction.