A q-RASAR approach for oral and inhalational toxicity prediction of perfluorinated and polyfluorinated compounds (PFCs) using rodent toxicity data

The global demand for perfluorinated and polyfluorinated compounds (PFCs), including the subclass of per- and polyfluoroalkyl substances (PFASs), has grown due to their extreme stability and resistance to heat, enabling diverse industrial applications. However, their environmental persistence and potential health risks have earned some the title ‘forever chemicals’. Regulations have been imposed to limit the use and release of PFCs into the environment. Computational studies offer a great alternative to animal studies, enabling broader, faster, and effective risk assessment while also being cost-effective. This study reports the development and validation of Quantitative Read-Across Structure-Activity Relationship (q-RASAR) models for the rodent toxicity of PFCs. This approach predicts toxicity for new compounds based on their structural resemblance to chemicals with known toxicity profiles. We employed a Partial Least Squares (PLS) regression algorithm to predict acute toxicity using oral pLD 50 and inhalation pLC 50 endpoints in both rat and mouse models. Our models demonstrated substantially improved predictive performance compared to previous research, with notably higher Q 2 F1 (an external validation metric) values across all four datasets: 0.969 for pLD 50 rats, 0.867 for pLD 50 mice, 0.917 for pLC 50 rats, and 0.938 for pLC 50 mice. This work addresses important gaps in the toxicity assessment of PFCs by providing models with more reliable predictions while adhering to the 3Rs principle (Replacement, Reduction, Refinement) of animal testing. The developed models were utilized to predict the toxicity of a true external set of commercially relevant PFCs, classifying previously uncharacterized compounds as potentially toxic or non-toxic, aiding future risk prioritization efforts.