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

Per- and polyfluoroalkyl substances (PFAS) are highly persistent synthetic chemicals that are widely detected in environmental and biological systems and are increasingly targeted by global regulatory actions. Their structural diversity, together with multiple and evolving PFAS definitions, complicates automated identification and classification across large chemical data sets. Here we present PFASGroups, an open-source cheminformatics framework for automated identification, structural classification, and prioritization of PFAS, designed for seamless integration into machine learning (ML) workflows. PFASGroups combines SMARTS-based functional group detection with graph-based analysis of per- and polyfluorinated components to assign PFAS groups and characterize their size, topology, and structural context. The framework implements an extensible library of PFAS group definitions, including OECD-derived classes, common organic functional groups, and fluorotelomer-specific patterns, and supports evaluation of multiple regulatory PFAS definitions (OECD, EU, OPPT 2023, UK, PFASSTRUCTv5). Across benchmark data sets, PFASGroups demonstrates computational performance suitable for large-scale screening and shows high agreement with existing tools, with systematic differences reflecting broader structural coverage and multigroup assignment. The graph-based representation enables robust handling of complex and highly branched molecules while maintaining high specificity in group detection. PFASGroups' identification and descriptor generation capabilities are readily usable in ML pipelines and for automatic prioritization. The module produces structural embeddings that integrate functional and graph descriptors, thereby improving predictive modeling of toxicological end points compared to conventional PFAS-specific fingerprints. Application to large chemical inventories, including the ECHA Classification and Labeling database, demonstrates its utility for high-throughput PFAS screening and regulatory decision support. PFASGroups is available as open-source software via GitHub (https://github.com/lucmiaz/PFASGroups) and PyPI, with a complementary browser-based interface for rapid PFAS screening.