Prediction of hepatotoxicity of traditional Chinese medicine compounds by support vector machine approach

In this study, based on literatures and web databases, 490 hepatotoxic compounds and 598 non-hepatotoxic compounds were selected as a data set for hepatotoxicity discriminative model generation. 1664 molecular descriptors, including physicochemical, charge distribution and geometrical descriptors, were calculated to characterize the molecular structure of liver toxic compounds. The combination of CfsSubsetEval valuation and BestFirst searching was used to choose molecular descriptors for model construction. With the help of support vector machine (SVM), a discriminative model with high accuracy was built. Meanwhile, the accuracy, sensitivity and specificity of this model were all above 80%. Besides, 23 traditional Chinese medicine compounds with hepatotoxicity were regarded as external validation, so as to further verify the model accuracy. Then, the present model was utilized to identify hepatotoxic compounds in Qingkailing injection. The results demonstrated that present study provides a reliable utility for the hepatotoxic compounds prediction in Chinese Medicinal Materials studies.