Dynamic hydrophobicity representation for predicting and interpreting ionizable organic micropollutant rejection by polyamide membranes

Molecular hydrophobicity is an essential property affecting solute-membrane interfacial affinity and, consequently, the rejection of organic micropollutants (OMPs) by polyamide nanofiltration and reverse osmosis membranes. In current mechanistic interpretations and machine-learning-based predictions of OMP rejection, LogD, a bulk-phase hydrophobicity descriptor at a specified pH, has been commonly employed. However, it cannot capture changes in hydrophobicity-related behaviors of ionizable OMPs as they migrate from bulk solution to the membrane interface. Here, we introduced a descriptor termed the ionization-induced hydrophobicity shift (IHS = LogP - LogD), and combined it with LogP, which represents the intrinsic hydrophobicity of the neutral molecular skeleton, to construct a dynamic hydrophobicity representation for OMP rejection modeling. Compared with LogD alone, the combined inputs of LogP and IHS improved prediction of XGBoost models, with the main benefit derived from ionized rather than neutral OMPs. Feature contribution and interaction analyses showed that IHS made a non-negligible contribution to predicted OMP rejection, while its effect varied with molecular size, charge state, and LogP. Analyses of molecular characteristics and interfacial interactions further supported two rejection-enhancement pathways related to OMP hydrophobicity: hydrophobic interfacial retention for weakly hydrated molecules with extended nonpolar or aromatic structures, and hydration-related transport resistance for highly polar or ionic molecules. Through decoupling intrinsic molecular hydrophobicity from ionization-induced shifts, this work provides a physically reasonable framework for better predicting and interpreting OMP rejection across polyamide membranes.