책임운영기관 농촌진흥청 국립축산과학원 (로고)

Predicting hepatotoxicity is an important component of safety-related evaluations of drug-like compounds. Hepatotoxicity is related to the physicochemical properties of drug-like compounds, especially their structural alerts. In this study, we developed a Bayesian model to predict the hepatotoxicities of 498 drug-like compounds based on their quantitative structure-toxicity relationships (QSAR). The devised model predicted the hepatotoxicity of these compounds using 25 structural descriptors (such as the ECFP6 fingerprint) and provided a sensitivity, specificity, and concordance of 97.2%, 86.9%, and 90.6%, respectively. The model also successfully classified the 498 drug-like compounds by hepatotoxicity. In addition, TOPKAT@ toxicity models were used to predict hepatotoxic effects related to toxicological endpoints (i.e., LD50, LOAEL) and liver injury-related potentials, such as Ames mutagenicity, carcinogenicity in male and female mice, and developmental toxicity potentials. Our results indicate that combined use of the devised Bayesian hepatotoxicity prediction model and the TOPKAT@ hepatotoxicity model could replace experimental safety assessments of drug-like compounds. Accordingly, adopting the devised Bayesian toxicity model would reduce the time and cost of animal-based toxicity research by considering the safety and effectiveness of candidate compounds.