Using Machine Learning to Classify Bioactivity for 3486 Per- and Polyfluoroalkyl Substances (PFASs) from the OECD List

Abstract

A recent OECD report estimated that more than 4000 per- and polyfluorinated alkyl substances (PFASs) have been produced and used in a broad range of industrial and consumer applications. However, little is known about the potential hazards (e.g., bioactivity, bioaccumulation, and toxicity) of most PFASs. Here, we built machine-learning-based quantitative structure–activity relationship (QSAR) models to predict the bioactivity of those PFASs. By examining a number of available molecular data sets, we constructed the first PFAS-specific database that contains the bioactivity information on 1012 PFASs for 26 bioassays. On the basis of the collected PFAS data set, we trained 5 different machine learning models that cover a variety of conventional models (e.g., random forest and multitask neural network (MNN)) and advanced graph-based models (e.g., graph convolutional network). Those models were evaluated based on the validation data set. Both MNN and graph-based models demonstrated the best performance. The average of the best area-under-the-curve score for each bioassay is 0.916. For predictions on the OECD list, most of the biologically active PFASs have perfluoroalkyl chain lengths less than 12 and are categorized into fluorotelomer-related compounds and perfluoroalkyl acids and their precursors.

Strategies for grouping per- and polyfluoroalkyl substances (PFAS) to protect human and environmental health

“…There are clearly relationships between PFAS structural elements and properties and behaviour (e.g. number of uorinated carbons in the peruoroalkyl(ether) chain, protein binding affinities, bioaccumulation potential, elimination rates, bioactivities within the PFAA/peruoroalkylether acid subclasses), 11,44,90,107 but on the other hand, critical toxic endpoints, as well as modes and mechanisms of action vary within the PFAS and such inconsistencies could limit the applicability of QSARs and thus reliability of computational tools.…”

Structure-Based Virtual Screening of Perfluoroalkyl and Polyfluoroalkyl Substances (PFASs) as Endocrine Disruptors of Androgen Receptor Activity Using Molecular Docking and Machine Learning

“…The 23 PFASs which have the potential to bind to the ligand binding site of AR are given in Table 3. Two of these compounds which we have predicted as active, 2,2,3,3,4,4,5,5,pentanamide (CASRN:150333-62-3) and 2,2,3,3,4,4,5,5,6,6,7,7,7-Tridecafluoro-N-(4-nitrophenyl)heptanamide (CASRN : 164796-44-5), have also been predicted by others as potential candidates for biological activity using machine learning, (Cheng and Ng, 2019) which further suggests that these chemicals are potential EDCs.…”

“…Several studies suggest that PFASs may be endocrine disrupting chemicals (EDCs) (Cheng and Ng, 2019;Parolini et al, 2016;Patlewicz et al, 2019;Sznajder-Katarzynska et al, 2019). EDCs are exogenous chemicals that interfere with hormone action, thereby increasing the risk of adverse health outcomes including cancer, reproductive impairment, cognitive deficits and obesity (La Merrill et al, 2020).…”

A Database Framework for Rapid Screening of Structure-Function Relationships in PFAS Chemistry

“…As the classification of PFAS is constantly improving, an automated PFAS classification system that can reflect the updates in PFAS classification rules is needed. Machine learning approaches have found use in identifying patterns in the existing data of PFAS's properties including bioactivity, bond strength, and sources and make predictions [16][17][18] . However, most of the machine learning methods in these studies are supervised learning which requires the molecules' structural information as features and properties as labels, whereas the number of PFASs with known properties is significantly lower than the number of PFASs with identified structures 13 .…”