人工神经网络用于苯砜基羧酸酯定量结构-活效关系研究

Study on QSAR of phenylsulfonyl carboxylate compounds by using artificial neural network

采用误差反传前向人工神经网络(artificial neural network,ANN)建立了56种苯砜基羧酸酯类化合物的结构与其对发光菌的急性毒性之间的定量关系模型(ANN模型).以56种苯砜基羧酸酯类化合物的量子化学参数作为输入,急性毒性作为输出,所构建网络模型的交叉检验相关系数为0.986 3、标准偏差为0.075 3、残差绝对值≤0.20,应用于外部预测集,预测集相关系数为0.988 0;而多元线性回归(multiple linear regression,MLR)法模型的相关系数为0.947 2、标准偏差为0.141 3、残差绝对值≤0.34.结果表明:ANN模型获得了比MLR模型更好的拟合效果.

The systematic study of quantitative structure-activity relationship （QSAR） on 56 phenylsulfonyl carbox ylate compounds between the structures and the acute toxicities to luminescent bacteria was performed by using the artificial neural network （ANN） based on the back propagation algorithm. For the ANN method, using the quantum chemical parameters about structure as the inputs and the acute toxicities as the outputs, the leave one out cross-val-idation regression coefficient was 0. 986 3, the standard error was 0. 075 3, the correlation coefficient of the test set was 0. 988 0 and the absolute values of residual were less than 0.20. In order to make contrast, the QSAR model was set up by using multiple linear regressions （MLR） method. For the model built by MLR, the correlation coeffi-cient was 0. 947 2, the standard error was 0. 141 3 and the absolute values of residual were less than 0.34. The re-sults showed that the performance of ANN method is better than that of MLR method.