Elucidating environmental fate and toxicological mechanisms of ultrashort- and short-chain PFAS: Integrating machine learning, molecular modeling, and experimental validation.

Per- and polyfluoroalkyl substances (PFAS), due to their recalcitrance, toxicity, and widespread environmental distribution, have emerged as a critical public health concern. The fate of ultrashort- and short-chain PFAS in the heterogeneous environment remains elusive. To address this, we employed a multiscale approach combining machine learning, density functional theory, molecular dynamics, and experimental validation. This framework enabled the prediction of critical environmental fate parameters, such as octanol-water and lipid-water partition coefficients, pK values, bioconcentration factors, and median lethal concentrations. These predictions demonstrated strong agreement with experimental data (R > 0.785). Using the machine learning-based PathDetect-SOM algorithm, we identified unique reorientation behaviors during membrane penetration, including partial recline, full recline, and oblique insertion. Unlike long-chain PFAS, ultrashort- and short-chain PFAS, particularly in their ionized forms, exhibit monolayer adsorption at lipid membranes, increasing the area per lipid and suggesting distinct toxicological mechanisms. Cytotoxicity assays and reactive oxygen species measurements further corroborated these findings, underscoring the environmental and health risks posed by short-chain PFAS. Structural and toxicological analyses, supported by Pearson correlation metrics and 26 molecular descriptors, revealed that the ionization state of PFAS significantly influences membrane uptake, whereas pK values showed no direct correlation with cytotoxicity. This study establishes a comprehensive framework for predicting the environmental fate and bioaccumulation potential of emerging contaminants, offering critical insights into their behavior in heterogeneous environments.