Fundamental Toxicological Sciences

Paper Details

Fundamental Toxicological Sciences
Vol. 6 No. 8 December 11, 2019 p.327-332
Original Article
Development of a hepatotoxicity prediction model using in vitro assay data of key molecular events
  • Koji Jojima (Division of Risk Assessment, National Institute of Health Sciences / Graduate School of Pharmaceutical Sciences, Osaka University / Present address: Chemicals Assessment and Research Center, Chemicals Evaluation and Research Institute, Japan / jojima-koji@ceri.jp)
Koji Jojima 1) 2) 3) , Takashi Yamada 1) , Akihiko Hirose 1) 2)
1) Division of Risk Assessment, National Institute of Health Sciences , 2) Graduate School of Pharmaceutical Sciences, Osaka University , 3) Present address: Chemicals Assessment and Research Center, Chemicals Evaluation and Research Institute, Japan
Keywords: Risk assessment, Screening tool, Hepatotoxicity, Repeated-dose toxicity
Abstracts

In this study, we developed screening-level hepatotoxicity prediction models using test data on in vitro assays, which measure key events at molecular levels that are possibly linked to hepatotoxicity. Hepatotoxic chemicals were retrieved from repeated-dose toxicity databases of the Hazard Evaluation Support System Integrated Platform and the Toxicogenomics Project. In vitro assay data with specified protein targets likely leading to hepatotoxicity were selected using the hepatotoxic chemicals. In total, 47 in vitro assays were selected for constructing the hepatotoxicity prediction models. Then, two predictive models were constructed. Model A returns “Hepatotoxic” if the query chemical is tested, and the test result is “Active” in any of the selected in vitro assays. Model B returns “Hepatotoxic” if an analog of the query chemical is tested, and the test result is “Active” in any of the selected in vitro assays. External validation of the two models was performed using repeated-dose toxicity test data from the Toxicity Reference Database. Model A and Model B had sensitivity values of 0.67 and 0.72 and specificity values of 0.74 and 0.72, respectively. Our models could predict the hepatotoxic chemicals underlying the toxic mechanisms that are not established by the existing knowledge base model. On the other hand, false negatives were found to involve mechanisms requiring metabolic activation. Because our hepatotoxicity prediction model is based on the biological activity of key molecular events leading to the toxicity endpoint, scientific justification would be more acceptable as adverse outcome pathway information becomes more available.