Hazard Prioritization and Transcriptomic Characterization of PFASs Based on Machine Learning and Human Liver Spheroids

Abstract Because of their pervasiveness and ongoing toxicity, particularly to the liver, perfluorinated and polyfluoroalkyl substances (PFASs) present an increasingly serious threat to human health and the environment. On the basis of high-throughput bioactivity data, we developed a quantitative structure–activity relationship (QSAR) model based on the XGBoost algorithm and an operational four-tier scheme for bioactivity-based prioritization of PFAS hepatotoxicity-related hazards. Public transcriptomic data from human liver spheroids (GSE145239) exposed to 19 PFAS compounds were subsequently analyzed to provide biological context for the model-derived hazard prioritization tiers. Transcriptomic response patterns differed among PFASs assigned to different hazard tiers. Transcriptomic profiling further suggested that lipid metabolism disruption and cellular stress responses were among the major biological processes associated with PFAS-induced hepatic perturbations. FABP1, PLIN2, and GDF15 were identified as shared responsive genes across PFAS exposure groups. This integrated computational and in vitro framework links molecular hazard prioritization to transcriptomic responses, providing a potential strategy for prioritizing data-scarce PFAS.