Modeling PFAS Sorption in Soils Using Machine Learning

Area of Science:

• Environmental Chemistry
• Soil Science
• Computational Chemistry

Background:

• Per- and polyfluoroalkyl substances (PFAS) are persistent environmental contaminants.
• Understanding PFAS sorption in soils is critical for risk assessment.
• Existing models for predicting PFAS sorption coefficients (Kd) have limitations.

Purpose of the Study:

• To develop and validate a novel machine learning (ML) tool, PFASorptionML, for predicting PFAS solid-liquid distribution coefficients (Kd) in soils.
• To identify key PFAS and soil properties influencing sorption behavior.
• To provide a user-friendly platform for environmental risk assessment.

Main Methods:

• Developed PFASorptionML using a dataset of 1,274 Kd entries for various PFAS in soils and sediments.
• Incorporated PFAS properties (molecular weight, hydrophobicity, pKa) and soil characteristics (pH, texture, organic carbon, CEC).
• Performed sensitivity analysis to determine the influence of different parameters on Kd values.

Main Results:

• PFASorptionML demonstrated high predictive performance with RPD > 3.16, outperforming existing tools.
• Molecular weight, hydrophobicity, and soil organic carbon content were identified as the most significant factors influencing PFAS sorption.
• PFAS chain length and functional group significantly impacted Kd, with longer chains and higher hydrophobicity increasing sorption.

Conclusions:

• PFASorptionML is a robust and accurate tool for predicting PFAS Kd in soils.
• The model's ability to integrate location-specific data enables spatial Kd mapping.
• PFASorptionML serves as a valuable resource for environmental risk assessment and management of PFAS contamination.