支持向量机用于酚类化合物毒性的QSAR研究

Application support vector machine to QSAR study of toxicity of substituted phe-nols

从分子结构出发,计算25个酚类化合物的分子连接性指数及分子的价连接性指数,用线性逐步回归方法建立4参数的最佳方程,以此4参数作为输入参数,将留一法(L00)应用到BP网络、径向基函数(RBF)神经网络,及新颖的机器学习方法支持向量机,建立酚类化合物预测黑呆头鱼毒性的QSAR模型。应用非线性SVM法建立的预测模型结果,优于BP网络和RBF网络,SVM、BP、RBF模型预测的相关系数分别为0.959,0.940和0.945,令人满意。

Multiple linear stepwise regression was used to investigated toxicity effect of a set of 25 substituted phenols on fathead minnows.The structural parameters are molecular connectivity indices.A best equation including four parameters was obtained.Then we established BPNNs,RBFNNs and SVM predictive models with four parameters above based on leave-one-out cross-validation method. The results showed that the SVM predictive model is better than others predictive model,the correlation coefficient of SVM,BPNNs and FBFNNs prediction model is 0.959,0.94 and 0.945,respectively.