摘要
采用BP神经网络模型研究了45种硝基芳烃类化合物的结构与其急性毒性之间的关系,以硝基芳烃类化合物的量子化学参数作为输入,用3×4×1网络预测其急性毒性。采用内外双重验证的办法分析和检验所得模型的稳定性,所构建网络模型的相关系数为0.999 5,交叉检验相关系数为0.996 8,标准差为0.023 5,残差绝对值≤0.15,应用于外部预测集,外部预测集相关系数为0.998 4;而多元线性回归法(MLR)模型的相关系数为0.943 5,交叉检验相关系数为0.928 7,标准差为0.240 9,残差绝对值≤0.69,外部预测集相关系数为0.956 6。结果表明,BP神经网络模型获得了比MLR模型更好的拟合效果。
The relationship between structure of 45 nitroaromatic compounds and its acute toxicity was studied by BP neural network based on the back propagation algorithm.For the BP neural network method,when using the quantum chemical parameters as the inputs of the neural network and the acute toxicity as the outputs of the neural network,the correlation coefficient of established model was 0.999 5,the leave one out cross-validation regression coefficient was 0.996 8,the standard error was 0.023 5,the correlation coefficient of the test set was 0.998 4 and the absolute values of residual were less than 0.15.In order to make a comparison,the QSAR model was set up by multiple linear regressions(MLR) method.For the model built by MLR,the correlation coefficient was 0.943 5,the leave one out cross-validation regression coefficient was 0.928 7,the standard error was 0.240 9 and the absolute values of residual were less than 0.69,the correlation coefficient of the test set was 0.956 6.The results showed that the performance of BP neural network method is better than that of MLR method.
出处
《湖北农业科学》
北大核心
2013年第5期1174-1176,1180,共4页
Hubei Agricultural Sciences
基金
河南省教育厅自然科学研究计划项目(2009B150023)
许昌市科技计划项目(5007)
许昌学院校内科研基金项目(2013067)
