QSAR Study on the Toxicity of Phenols for Fathead Minnows by Using Support Vector Machine and Neural Networks

Support victor machine (SVM) and artificial neural networks (ANNs) including back-propagation network (BPNN) and radial basis function network (RBFNN) were used to investigate toxic effect of phenols on fathead minnows. Molecular connectivity index was used as structural descriptor. The applicability of established BPNNs, RBFNNs and SVM models based on optimized parameters was compared using leave-one-out (LOO) cross-validation method. Results showed that all the models investigated were applicable for the quantitative structure-activity relationship (QSAR) studies and the SVM model is slightly better than others. The correlation coefficients between measured toxicities and predicted values of SVM, BPNN and RBFNN models are 0.959, 0.94, and 0.945, respectively.