4311696

Integration of transfer learning and multitask learning to predict the potential of per- and polyfluoroalkyl substances in activating multiple nuclear receptors associated with hepatic lipotoxicity

Date
August 18, 2025

Per- and polyfluoroalkyl substances (PFAS) are known to activate nuclear receptors (NRs), which leads to hepatic lipotoxicity. Given the extensive variety of PFAS and the broad spectrum of their NR targets, machine learning (ML) models are appealing tools for predicting their activities on NRs. However, the limited availability of PFAS data poses challenges for training conventional ML models to generate accurate predictions. To overcome this limitation, we evaluated five conventional ML models trained on three datasets with varying chemical diversity, and explored a transfer learning (TL) strategy combined with either a single-task (ST) or a multitask (MT) deep neural network (DNN), to predict PFAS-induced activation of five NRs associated with hepatic lipotoxicity. The TL-MT-DNN outperformed the TL-ST-DNN and the conventional ML models in both discrimination and positive class recognition, demonstrating its strong capability to effectively transfer knowledge from related chemical domains to address PFAS data scarcity. This model was subsequently applied to 3716 PFAS from the PFASSTRUCTv5 database, among which 230 PFAS potentially activate all five NRs. To validate the model performance, ten representative PFAS were individually dosed in L-02 cells to measure their agonistic activities on five NRs. The results verified that the established TL-MT-DNN model produced reliable NR activation predictions, particularly for PPARα and PPARγ.

Related Products

Thumbnail for Comprehensively unveiling the human exposure risk to halogenated organic compounds through non-target analysis
Comprehensively unveiling the human exposure risk to halogenated organic compounds through non-target analysis
The restriction on legacy halogenated organic compounds (HOCs) has given rise to a more intricate situation of human exposure owing to the diverse range of replacement chemicals…
Thumbnail for Comprehensive screening of Per- and polyfluoroalkyl substances in cosmetics and personal care products: Implication for human dermal exposure risk
Comprehensive screening of Per- and polyfluoroalkyl substances in cosmetics and personal care products: Implication for human dermal exposure risk
Per- and polyfluoroalkyl substances (PFAS) in beauty products pose a potential health risk through dermal contact for the general public…