Context-Aware Hydrophobicity Modeling: HydroMap and FastHydroMap

Area of Science:

• Biophysics
• Computational Chemistry
• Materials Science

Background:

• Hydrophobicity is crucial for molecular interactions and assembly, traditionally viewed as an additive property.
• Current methods for quantifying hydrophobicity (dewetting free energy, Fdewet) via molecular simulation are computationally expensive.
• Existing sequence-based hydropathy scales oversimplify hydrophobicity, neglecting collective surface effects.

Purpose of the Study:

• To develop computationally inexpensive models for predicting residue-level dewetting free energy (Fdewet).
• To capture context-dependent hydrophobicity by considering local water features and collective surface properties.
• To enable rapid scoring for materials design and analysis of dynamic hydrophobic processes.

Main Methods:

• Extracted local water features (structural signatures, residue-water potential energy) from brief all-atom simulations.
• Developed HydroMap model to predict Fdewet directly from these water features.
• Created FastHydroMap, a graph neural network surrogate trained on HydroMap, requiring no solvent simulation.

Main Results:

• Residue-level Fdewet can be accurately predicted from local water features.
• HydroMap and FastHydroMap capture context-dependent hydrophobicity, outperforming classical scales.
• Applied models to α-synuclein, calmodulin, and Protein G, revealing hidden binding sites, dynamic changes upon Ca2+ binding, and folding trajectories.

Conclusions:

• HydroMap and FastHydroMap provide a computationally efficient way to quantify hydrophobicity.
• These models enable practical, physically grounded design of hydrophobic interactions.
• The approach facilitates rapid materials design and time-resolved analysis of hydrophobic-mediated biological processes.