Characterizing And Predicting Hydrophobicity At The Nanoscale To Inform Protein Interactions

Hydrophobicity - the aversion of liquid water to surfaces - is a key property property of proteins, underlying diverse processes from intermolecular interactions, to phase behavior, to protein folding. However, rigorous characterization of the hydrophobicity at the nanoscale of topographically and chemically complex solutes such as biomolecules remains an unsolved problem. While numerous approximations, such as hydropathy scales based on single residue properties, exist, they fail to account for the complex, many-bodied response of proximal waters to a protein that underly its emergent hydrophobicity. Here, we apply a combination of Molecular Dynamics (MD) simulations with the Indirect Umbrella Sampling (INDUS) method to rigorously capture and quantify the hydrophobicity of complex solutes such as proteins. We find that protein hydration waters are susceptible to unfavorable energetic perturbations. Furthermore, we use these findings to identify the most hydrophobic regions of the surfaces of a variety of proteins. We find that, contrary to conventional treatments and understanding of hydrophobicity, these hydrophobic regions consist of similar numbers of non-polar and polar surface atoms. Furthermore, we test the correspondence between these hydrophobic surface regions and experimentally determined protein-protein interaction interfaces. We find a striking correspondence between the hydrophobic regions uncovered from our simulations and known protein interaction interfaces, suggesting that the indirect contribution of hydrophobic effects play a significant role in protein recognition across a variety of proteins. Finally, we combine state-of-the-art machine learning (ML) methodologies with INDUS to construct a series of models to predict the hydrophobicity of self-assembled monolayer (SAM) chemically heterogeneous surfaces. As sophisticated ML models tend to operate as `black-boxes', obscuring interpretability, we use the tools of explainable artificial intelligence (XAI) to arrive at a model of surface hydrophobicity that is at once extremely accurate and physically interpretable. This model suggests an interesting asymmetry in the influence of additional polar or non-polar groups on the hydrophobicity of a surface.