DeepHeptox: An Interpretable Deep Learning Model for Multi-Endpoint Hepatotoxicity Prediction of Chemical Compounds

Le résumé fourni par la source

Hepatotoxicity represents a major adverse outcome of chemical exposure, as the liver plays a central role in xenobiotic metabolism and detoxification. Accurate prediction of hepatotoxicity is therefore essential for drug development and chemical safety assessment. Computational methods provide an efficient and ethical alternative for assessing hepatotoxicity before experimental validation. To address this need, we developed DeepHeptox, a deep learning model based on Graph Attention Networks (GAT) for multi-endpoint hepatotoxicity prediction, covering hepatitis, jaundice, elevated liver enzymes, hepatocellular injury, hepatic fibrosis, hepatomegaly, and cholestasis. DeepHeptox achieved area under the ROC curve (AUC) values exceeding 0.87 and accuracy (ACC) above 0.80 for both overall and endpoint-specific predictions on the test set. Our approach enables both the identification of structural alerts (SAs) via Klekota-Roth fingerprint (KRFP) analysis and the visualization of molecular substructure importance for hepatotoxicity predictions through GNNExplainer. A user-friendly web server is provided for interactive prediction and visualization. DeepHeptox offers a practical tool for early stage toxicity screening in drug development and chemical safety evaluation.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.