Modeling PFAS sorption in soils using machine learning

Per- and polyfluoroalkyl substances (PFAS) are emerging pollutants of global environmental concern due to their persistence, widespread occurrence, and toxicity. Accurate PFAS sorption data in soils is essential for assessing their fate and transport in the environment; however, current prediction models often lack precision and broader applicability. To address this limitation, we present PFASorptionML, an advanced machine learning (ML)-based tool designed to predict the solid-liquid distribution coefficients (K_d) of 49 PFAS compounds with diverse chemical structures, including ionizable PFAS with environmentally relevant acid dissociation constants (pKa), in soils.

We developed an extensive literature-based sorption dataset comprising 1,274 K_d (PFAS) entries across 47 peer-reviewed studies. This dataset enabled a critical evaluation of the effects of PFAS chain length and functional groups on sorption behavior. This dataset was used to train the ML model, which integrates PFAS-specific properties—such as molecular weight, hydrophobicity, and charge density—with soil-specific properties, including pH, organic carbon content, texture, and cation exchange capacity. Before training the model, gaps in soil property data were addressed using advanced imputation techniques (e.g., K-nearest neighbor), ensuring data completeness and reliability. Sensitivity analysis revealed the dominant role of hydrophobic interactions and the minor contribution of electrostatic interactions in PFAS sorption, highlighting the importance of incorporating these factors into environmental modeling.

Beyond its predictive capabilities, PFASorptionML represents a significant advancement in PFAS modeling for environmental scenarios. It enables the generation of high-resolution European K_d (PFAS) maps by integrating soil property repositories (e.g., LUCAS EU dataset), thereby upscaling laboratory findings to European conditions. Furthermore, PFASorptionML offers a free-to-use online platform for practitioners, supporting risk assessment, groundwater management, and the development of effective remediation strategies for PFAS-contaminated sites.