Using machine learning to classify the immunosuppressive activity of per- and polyfluoroalkyl substances

Area of Science:

• Environmental Toxicology
• Computational Chemistry
• Immunotoxicology

Background:

• Per- and polyfluoroalkyl substances (PFASs) are persistent organic pollutants with known immunosuppressive effects.
• Evaluating PFAS immunosuppression is crucial for regulatory toxicology and risk assessment.
• Existing methods for assessing PFAS immunotoxicity are limited in scope and efficiency.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting the immunosuppressive activity of PFASs.
• To identify key molecular descriptors associated with PFAS immunosuppressive effects.
• To establish a framework for efficient assessment of a large number of PFAS compounds.

Main Methods:

• Collected data on 146 PFASs, including their structures and concentration gradients, from scientific literature.
• Utilized Dragon descriptors for structural characterization and employed feature selection techniques (importance analysis, stepwise elimination).
• Developed and compared three ML models: Random Forest (RF), Extreme Gradient Boosting Machine (XGB), and Categorical Boosting Machine (CB).

Main Results:

• All three ML models demonstrated excellent predictive performance, with the RF model achieving an average AUC of 0.9720 on the testing set.
• Feature importance analysis identified concentration, SpPosA_X, IVDE, R2s, and SIC2 as critical molecular features influencing immunosuppressive activity.
• Applicability domain analysis confirmed reliable prediction boundaries for the developed models.

Conclusions:

• This study represents the first application of ML models to investigate PFAS immunosuppressive activity.
• The identified molecular features provide insights into the mechanisms underlying PFAS immunotoxicity.
• The developed models offer a powerful tool for efficiently assessing the immunosuppressive potential of numerous PFASs, thereby informing environmental and health risk assessments.