PFASGroups: An Open-Source Framework for Automated Identification, Structural Classification, and Prioritization of...

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Toxicology

Background:

• Per- and polyfluoroalkyl substances (PFAS) are persistent synthetic chemicals found globally.
• Regulatory actions against PFAS are increasing due to environmental and health concerns.
• Automated identification and classification of diverse PFAS structures are challenging.

Purpose of the Study:

• To develop an open-source cheminformatics framework, PFASGroups, for automated PFAS identification and classification.
• To facilitate integration into machine learning workflows for large-scale chemical data analysis.
• To support regulatory decision-making and toxicological assessments of PFAS.

Main Methods:

• PFASGroups combines SMARTS-based functional group detection with graph-based analysis.
• It characterizes PFAS size, topology, and structural context.
• The framework supports multiple regulatory PFAS definitions and includes an extensible library of group definitions.

Main Results:

• PFASGroups demonstrates computational performance suitable for high-throughput screening.
• It shows high agreement with existing tools, with broader structural coverage.
• The graph-based representation effectively handles complex molecules and improves specificity.
• Generated structural embeddings enhance predictive modeling of toxicological endpoints.

Conclusions:

• PFASGroups provides a robust solution for automated PFAS identification, classification, and prioritization.
• Its integration capabilities benefit machine learning pipelines for environmental and toxicological studies.
• The framework is valuable for high-throughput screening of chemical inventories and regulatory support.