Using Machine Learning to Predict First-Order Reaction Rate Constants of PFAS Degradation

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Chemical Engineering

Background:

• Per- and polyfluoroalkyl substances (PFAS) are persistent environmental pollutants requiring efficient remediation methods.
• Predicting PFAS degradation kinetics is crucial for developing effective treatment technologies.

Purpose of the Study:

• To develop a machine learning (ML) framework for predicting the first-order reaction rate constant (k) of PFAS degradation.
• To identify key molecular and experimental factors influencing PFAS degradability across various remediation processes.

Main Methods:

• Utilized Extreme Gradient Boosting (XGB), Random Forest (RF), Extra Trees (ET), and Gradient Boosted Regression Trees (GBRT) models.
• Integrated molecular descriptors (MD) and experimental conditions (reaction type, initial concentration) as input features.
• Applied the best-performing XGB model to predict degradation rates for 2,631 PFAS from the OECD database.

Main Results:

• The XGB model demonstrated the best predictive performance (R² = 0.568, RMSE = 0.448).
• Experimental conditions, particularly reaction type and initial PFAS concentration, significantly influenced degradation rates more than molecular features.
• Clustering analysis identified PFAS structures with higher degradation potential, including those with C-Br bonds and -SO₃H groups.

Conclusions:

• The developed ML framework provides a scalable approach for assessing PFAS degradability.
• Understanding the influence of experimental parameters is vital for optimizing PFAS remediation strategies.
• Structural insights can guide the selection of PFAS for targeted degradation efforts.