Structure-resolved adsorption of PFAS on humic substances: A combined study of molecular dynamics, explainable machine learning, and experiments

An MD–spectroscopy–XAI framework was developed for PFAS–HS interactions • Ensemble ML accurately predicted PFAS adsorption across 410 structures • Adsorption strength increased while diffusivity decreased with PFAS size • PFAS adsorption energy negatively correlated with molecular weight ( r  = −0.72) • SHAP identified MolMR, Kappa3, size, and flexibility as key driver Understanding the adsorption behavior of PFAS on humic substances (HS) is critical for predicting their environmental fate and developing effective remediation strategies. Here, we integrate molecular dynamics (MD) simulations, experimental characterization, and explainable machine learning (XAI) to resolve PFAS interactions with HS at the molecular level. Based on 410 PFAS structures, MD-derived adsorption energies reveal a clear structure-dependent adsorption trend, with a strong negative correlation (Pearson's |r| > 0.72) between molecular weight and adsorption energy magnitude. Machine learning models show robust predictive performance (R 2  > 0.88), enabling accurate adsorption-energy prediction across structurally diverse PFAS. SHAP analysis reveals that PFAS adsorption onto HS is governed by a synergistic interplay of multiple molecular descriptors rather than any single molecular property. Molar refractivity, Kappa3, Number of hydrogen bond acceptors, and molecular weight emerge as the most influential descriptors, highlighting the coupled roles of electronic polarizability, molecular topology, conformational adaptability, and size in controlling adsorption. In contrast, descriptors associated with polarity and hydration tendency are more closely linked to migration behavior. Adsorption experiments and complementary spectroscopic and microscopic characterizations consistently confirmed stronger fluorine accumulation and greater compaction of the HS matrix induced by long-chain PFAS. Overall, the results converge on a unified molecular picture in which larger, more polarizable, and structurally adaptable PFAS are preferentially retained, whereas smaller and more hydration-favored PFAS remain comparatively mobile. This work provides a physics-informed and interpretable framework for predicting PFAS transport in heterogeneous organic environments. Download: Download high-res image (229KB) Download: Download full-size image