In silico prediction of mitochondrial toxicity by using GA-CG-SVM approach

Abstract

Drug-induced mitochondrial toxicity has become one of the key reasons for which some drugs fail to enter market or are withdrawn from market. Thus early identification of new chemical entities that injure mitochondrial function grows to be very necessary to produce safer drugs and directly reduce attrition rate in later stages of drug development. In this study, support vector machine (SVM) method combined with genetic algorithm (GA) for feature selection and conjugate gradient method (CG) for parameter optimization (GA-CG-SVM), has been employed to develop prediction model of mitochondrial toxicity. We firstly collected 288 compounds, including 171 MT+ and 117 MT-, from different literature resources. Then these compounds were randomly separated into a training set (253 compounds) and a test set (35 compounds). The overall prediction accuracy for the training set by means of 5-fold cross-validation is 84.59%. Further, the SVM model was evaluated by using the independent test set. The overall prediction accuracy for the test set is 77.14%. These clearly indicate that the mitochondrial toxicity is predictable. Meanwhile impacts of the feature selection and SVM parameter optimization on the quality of SVM model were also examined and discussed. The results implicate the potential of the proposed GA-CG-SVM in facilitating the prediction of mitochondrial toxicity.