A New CSRML Structure-Based Fingerprint Method for Profiling and Categorizing Per- and Polyfluoroalkyl Substances (PFAS)

Abstract

The term PFAS encompasses diverse per- and polyfluorinated alkyl (and increasingly aromatic) chemicals spanning industrial processes, commercial uses, environmental occurrence, and potential concerns. With increased chemical curation, currently exceeding 14,000 structures in the PFASSTRUCTV5 inventory on EPA’s CompTox Chemicals Dashboard, has come increased motivation to profile, categorize, and analyze the PFAS structure space using modern cheminformatics approaches. Making use of the publicly available ToxPrint chemotypes and ChemoTyper application, we have developed a new PFAS-specific fingerprint set consisting of 129 TxP_PFAS chemotypes coded in CSRML, a chemical-based XML-query language. These are split into two groups, the first containing 56 mostly bond-type ToxPrints modified to incorporate attachment to either a CF group or F atom to enforce proximity to the fluorinated portion of the chemical. This focus resulted in a dramatic reduction in TxP_PFAS chemotype counts relative to the corresponding ToxPrint counts (averaging 54%). The remaining TxP_PFAS chemotypes consist of various lengths and types of fluorinated chains, rings, and bonding patterns covering indications of branching, alternate halogenation, and fluorotelomers. Both groups of chemotypes are well represented across the PFASSTRUCT inventory. Using the ChemoTyper application, we show how the TxP_PFAS chemotypes can be visualized, filtered, and used to profile the PFASSTRUCT inventory, as well as to construct chemically intuitive, structure-based PFAS categories. Lastly, we used a selection of expert-based PFAS categories from the OECD Global PFAS list to evaluate a small set of analogous structure-based TxP_PFAS categories. TxP_PFAS chemotypes were able to recapitulate the expert-based PFAS category concepts based on clearly defined structure rules that can be computationally implemented and reproducibly applied to process large PFAS inventories without need to consult an expert. The TxP_PFAS chemotypes have the potential to support computational modeling, harmonize PFAS structure-based categories, facilitate communication, and allow for more efficient and chemically informed exploration of PFAS chemicals moving forward.

Introduction to the Special Issue: AI Meets Toxicology

“…The methods described in this special issue cover a wide range of AI methods ranging from expert systems, ,, over similarity measures including read-across methods, − to classical machine learning such as random forests (RF), support vector machines (SVM), and artificial neural networks (ANN) − ,, to deep learning (DL) methods ,,− , including equivariant neural networks, deep generative models, and even large language models . In addition to models relying purely on the chemical structure, there is a notable trend of bringing in additional modalities to improve or inform predictive models. , In the following, we provide an overview of the AI approaches used in the publications contained in the SI.…”

“…The tasks and problems spanned by the proposed methods are also highly diverse and range from classical quantitative structure–activity/property relationship (QSAR/QSPR) problems, such as the prediction of drug-induced liver injury (DILI), , Ames mutagenicity, , or acetylcholinesterase (AChE) inhibition, to categorization of chemicals, generating transcriptomic profiles and classifying parts of regulatory documents . In their paper, “Transparency in Modeling through Careful Application of OECD’s QSAR/QSPR Principles via a Curated Water Solubility Data Set”, the authors revisited the five Organisation for Economic Co-operation and Development (OECD) principles for QSAR models.…”

High Throughput Read-Across for Screening a Large Inventory of Related Structures by Balancing Artificial Intelligence/Machine Learning and Human Knowledge

“…This methodology has been applied to assist regulatory grouping of industrial chemicals including organic flame retardants (OFRs) 34 and per/polyfluorinated substances (PFASs). 35 Recently, Adams et al also developed a new set of chemotypes for an analogue identification methodology for the readacross approaches in GenRA. 36 To achieve these goals for our use case of the bisphenol-202 set, bisphenol-specific chemotypes were designed based on the case of 15 identified clusters; they were mapped to clusters in Figure 2 (diagonal names at the base).…”

“…Designing of new structural categories is based on the CSRML (Chemical Subgraph and Reaction Mark-up Language) , technology extending beyond default ToxPrint chemotypes. This methodology has been applied to assist regulatory grouping of industrial chemicals including organic flame retardants (OFRs) and per/polyfluorinated substances (PFASs) . Recently, Adams et al.…”

Building Chemical Intuition about Physicochemical Properties of C8-Per-/Polyfluoroalkyl Carboxylic Acids through Computational Means

“…Within the last few decades, the scientific community has awakened to the threat that per- and polyfluorinated alkyl substances (PFAS) pose toward human and environmental safety. − These substances have been demonstrated to be immunopathogenic, carcinogenic, and in preliminary zebrafish studies, even teratogenic. ,, The daunting reality of this issue is that there exist over 8000 PFAS substances across the global market for a wide range of applications. − However, very few of these chemicals have been tested for toxicity, despite the demonstrated health risks of some PFAS, or have had their physicochemical properties quantified, despite the need to understand their fate and transport. One of the major goals set out by the PFAS research community is the attainment of physicochemical properties .…”