LAUSR

Significance Diverse biomolecular recognition and self-assembly processes are driven by hydrophobic interactions between surfaces that display nanoscale chemical heterogeneity. However, the relationship between chemical patterning and hydrophobicity is nontrivial and cannot be captured by additive approaches, such as hydropathy scales. Here, we combine molecular simulations and machine learning to learn this relationship and develop predictive models of hydrophobicity that are accurate, interpretable, and generalizable. The learned models unveil new insights into the chemical determinants of hydrophobicity and highlight the importance of accounting for chemical correlations between surface groups in determining hydrophobicity. Our models could spur the rational design of soft materials and biomolecules with optimal hydrophobicity/hydrophilicity.