Data-Driven Molecular Design of TFC Membrane Monomers with Reduced PFAS Affinity

Per- and polyfluoroalkyl substances (PFAS) are persistent aquatic contaminants that challenge nanofiltration (NF) and reverse osmosis (RO) treatment, particularly when trace-level permeation compromises water-quality goals. Beyond steric and electrostatic exclusion, PFAS transport through thin-film composite (TFC) selective layers can be influenced by solute-membrane affinity and partition-mediated diffusion. Here, we developed an integrated quantum-chemical and interpretable machine-learning framework to identify monomer-level features associated with weak PFAS binding. A data set of 5,265 complexes was constructed by pairing 39 TFC membrane-relevant monomers with 135 PFAS fragments. Density functional theory and energy decomposition analysis were used to quantify interaction energies, and a Bayesian-optimized LightGBM model captured the monomer-PFAS fragment interaction landscape. SHAP and causal discovery revealed a synergistic dual mechanism: global polarity matching and dipole polarization drive long-range approach, while local electrostatic anchoring between PFAS oxygens and monomer hydrogens dictates interfacial orbital overlap. Clustering separated weak-binding, fluorinated-tail-dominated complexes from stronger-binding, polar-headgroup-dominated complexes, highlighting the importance of PFAS functional groups in interfacial association. Cross-attention-based motif analysis identified nitrogen-containing aliphatic-chain structures as low-affinity candidates. Additional quantum-chemical validation with PFOA and PFOS supported their weak, dispersion-dominated interactions. This work provides molecular design guidance for TFC polymeric membranes with reduced PFAS sorption tendency.