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

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.

Formation of PFAAs in fish through biotransformation: A PBPK approach.

A physiologically-based pharmacokinetic (PBPK) model for perfluorinated alkyl acids (PFAAs) in rainbow trout has been updated to include formation of perfluorooctanoic acid (PFOA) from the biotransformation of 8:2 fluorotelomer carboxylic acid (8:2 FTCA). The updated model is dynamic and simulates both uptake and depuration phases. Two empirical studies are used to parameterize and test the model. In the first case, parameters related to fecal elimination and protein binding were optimized. In the second case, parameters were sourced either from literature or from optimized values based on the first study to test model performance. Optimization of parameters resulted in a decrease in the difference between experimental data and simulation results by 57 and 23 percent for the first and the second case, respectively, compared to the original case. Sensitivity analysis was performed to identify important parameters, and uncertainty in model prediction propagated by these parameters was assessed using Monte Carlo analysis. For each case, 80 and 89 percent, respectively, of median predicted values were within the limits of experimental error when comparing simulated and experimental data. This is the first toxicokinetic model that incorporates biotransformation of PFAA precursors and simultaneously predicts the distribution of the precursor and metabolite in different tissues. The model is mechanistic, and could be applied to simulate a variety of scenarios by using the organism-specific physiological properties compiled here with other chemical-specific parameters (e.g. protein interactions).

Evaluating parameter availability for physiologically based pharmacokinetic (PBPK) modeling of perfluorooctanoic acid (PFOA) in zebrafish.

Physiologically based pharmacokinetic (PBPK) models are considered useful tools to describe the absorption, distribution, metabolism and excretion of xenobiotics. For accurate predictions, PBPK models require species-specific and compound-specific parameters. Zebrafish are considered an appropriate vertebrate model for investigating the toxicity of a wide variety of compounds. However, no specific mechanistic model exists for the pharmacokinetics of perfluoroalkyl acids (PFAAs) in zebrafish, despite growing concern about this class of ubiquitous environmental contaminants. The purpose of this study was to evaluate the current state of knowledge for the parameters that would be needed to construct such a model for zebrafish. We chose perfluorooctanoic acid (PFOA) as a model PFAA with greater data availability. We have updated a previous PBPK model for rainbow trout to simulate PFOA fate in zebrafish following waterborne exposure. For the first time, the model considers hepatobiliary circulation. In order to evaluate the availability of parameters to implement this model, we performed an extensive literature review to find zebrafish-specific parameters. As in previous approaches, we broadened our search to include mammalian and other fish studies when zebrafish-specific data were lacking. Based on the method used to measure or estimate parameters, or based on their species-specific origin, we scored and ranked the quality of available parameters. These scores were then used in Monte Carlo and partial rank correlation analyses to identify the most critical data gaps. The liver, where fatty acid binding proteins (FABPs) and plasma proteins are considered, represented the best model-data agreement. Lack of agreement in other tissues suggest better parameters are needed. The results of our study highlight the lack of zebrafish-specific parameters. Based on sensitivity and uncertainty analysis, parameters associated with PFAA-protein interactions and passive diffusion need further refinement to enable development of predictive models for these emerging chemicals in zebrafish.

Socio-economic analysis for the authorisation of chemicals under REACH: a case of very high concern?

Under the European chemicals' legislation, REACH, substances that are identified to be of "very high concern" will de facto be removed from the market unless the European Commission grants authorisations permitting specific uses. Companies who apply for an authorisation without demonstrating "adequate control" of the risks have to show by means of a socio-economic analysis (SEA) that positive impacts of use outweigh negative impacts for human health and ecosystems. This paper identifies core challenges where further in-depth guidance is urgently required in order to ensure that a SEA can deliver meaningful results and that it can effectively support decision-making on authorisation. In particular, we emphasise the need (i) to better guide the selection of tools for impact assessment, (ii) to explicitly account for stock pollution effects in impact assessments for persistent and very persistent chemicals, (iii) to define suitable impact indicators for PBT/vPvB chemicals given the lack of reliable information about safe concentration levels, (iv) to guide how impacts can be transformed into values for decision-making, and (v) to provide a well-balanced discussion of discounting of long-term impacts of chemicals.