[Application of support vector machine approach in studying nephron toxicity of Chinese medicinal materials]

On the basis of web databases, 111 compounds with nephrotoxicity and 90 compounds without nephrotoxicity were collected as data set of nephrotoxicity discrimination model, 39 compounds with tubular necrosis and 39 compounds without tubular necrosis were collected as data set of tubular necrosis discrimination model. The 6 122 molecular descriptors, including physicochemical, charge distribution and geometrical descriptors were calculated to characterize the molecular structure of the above-mentioned compounds. CfsSubsetEval valuation method and BestFirst-D1-N5 searching method were used to select molecular descriptors. Two models with high accuracy were built based on the support vector machine (SVM) approach, respectively. Accuracy, sensitivity, specificity and matthew's correlation coefficient of the two models were all above 70%. By using 22 nephrotoxicity compounds of Chinese medicine, the nephrotoxicity discrimination model was further verified with an accuracy of 72.73%. Using the tubular necrosis discrimination model, 10 potential compounds which can cause tubular necrosis were screened from the positive results of nephrotoxicity discrimination model, 6 of them have been verified by literatures. The results demonstrated that the discrimination models can be applied to screen nephrotoxic compounds from Chinese medicinal materials, and they also offer a new research idea for the further studies on the mechanism of nephrotoxicity.