Physicochemical Property Models for Poly- and Perfluorinated Alkyl Substances and Other Chemical Classes

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Toxicology

Background:

• Per- and polyfluoroalkyl substances (PFAS) pose environmental risks due to data gaps in their physicochemical properties.
• Predictive models are crucial for assessing PFAS environmental fate, transport, and exposure.

Purpose of the Study:

• Develop and validate quantitative structure-property relationship (QSPR) models for key physicochemical properties of PFAS.
• Compare the performance of global QSPR models against local models and existing tools.

Main Methods:

• Extracted over 200,000 experimental property records from public data sources.
• Developed global QSPR models using diverse chemical classes and local models using PFAS-only data.
• Assessed model performance using R-squared and mean absolute error (MAE) on test data.
• Implemented a nearest neighbor-based applicability domain (AD) measure.

Main Results:

• Global models demonstrated superior prediction accuracy for PFAS properties compared to local models (11% MAE reduction).
• Global models achieved strong performance across property endpoints (R-squared 0.76-0.89).
• New models showed a 13% MAE reduction compared to existing Toxicity Estimation Software Tool (T.E.S.T) models.
• The applicability domain effectively excluded poor predictions, retaining ~95% of chemicals within the AD.

Conclusions:

• Global QSPR models provide accurate predictions for PFAS physicochemical properties, essential for environmental risk assessment.
• The developed models and applicability domain enhance the reliability of environmental exposure predictions for PFAS.
• These findings support better regulatory and management strategies for PFAS contamination.