ZHANG Xin-you, LI Ying, WU Di-yao, MIN Jie
Objective: To develop a quantitative structure-activity relationship (QSAR) liver toxicity prediction and evaluation method for medicinal and food homologous (MFH) substances, and apply it to the liver toxicity evaluation of Ginkgo biloba and Polygonum multiflorum. Methods: A total of 1 110 compounds with hepatotoxic adverse reactions were collected from the SIDER database (SIDER 4.1, released on October, 2015) and LiverTox (updated on March, 2025) as positive datasets, and 312 compounds that do not cause liver damage from DILIrank (updated on September, 2023) as negative datasets. Stone MIND Collector software was used to convert the collected compound molecule images into SMILES format, and then a hepatotoxicity model was constructed using Inno QSAR on the DrugFlow platform. The accuracy and robustness of the QSAR hepatotoxicity model were verified using 5-fold cross validation. Then, 72 compounds contained in Ginkgo biloba and 25 compounds contained in Polygonum multiflorum were collected and integrated from the TCMSP (Version 2.3, updated on May, 2014) database and Yaozhiwang database (updated on quarter 3rd) for screening. The QSAR hepatotoxicity model was used to predict and evaluate the hepatotoxicity of the two medicinal substances. Results: The fitting evaluation of the constructed QSAR hepatotoxicity model showed an ACC of 0.809, ROC-AUC of 0.757, close to 1. The F1 score value was 0.888, indicating that the model has high accuracy and precision, good model performance, and high predictability. The prediction results showed that both Ginkgo biloba and Polygonum multiflorum are substances that may cause liver toxicity. The hepatotoxicity caused by Ginkgo biloba is most likely related to the compounds cardanol (C21H34O), amentoflavone (C30H18O10), ginkgolide B (C20H24O10), and ginkgolide J (C20H24O10). The hepatotoxicity caused by Polygonum multiflorum is most likely related to the compounds pyrogallol (C6H6O3), 2,3,5,4'-tetrahydroxystilbene-2-O-β-D-glucoside (C20H22O9), and citreorosein (C15H10O6). Conclusion: The QSAR liver toxicity prediction and evaluation method for MFH substances established in this study aims to explore and evaluate the liver toxicity of food and drug substances more deeply and finely, providing data reference for early identification and warning.