用分子电性距离矢量和人工神经网络研究新型均三氮苯类衍生物的除草活性  

Investigation on the phytocidal activities of novel 1,3,5-triazine derivatives by electronegativity distance vector and artificial neural network

作　　者：陈艳[1] 冯长君[1] 洪涛[1] 

机构地区：[1]徐州工程学院化学化工学院,江苏徐州221111

出　　处：《计算机与应用化学》2014年第3期329-332,共4页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(21272095);徐州工程学院培育项目(XKY2011102)

摘　　要：以分子电性距离矢量（MEDV-13）表征新型均三氮苯类衍生物的分子结构，通过最佳变量子集回归建立了34种化合物除草活性的QSAR模型，模型的相关系数为0．888。模型通过RCV^2、VIF等指标检验具有良好的稳健性和预测能力。根据进入模型的3个电性距离矢量m15、m56、m91来看，影响除草剂除草活性的主要因素是分子的-CH2-、〉CH-、-N-和-X等结构片段。以m15、m56、m91为人工神经网络的输入层，设定3：6：1的网络结构，所建BP模型的相关系数为0．976，相关性明显高于多元线性回归模型。结果表明，用电性距离矢量表征均三氮苯类衍生物的除草活性是合理而有效的。Molecular electronegativity-distance vector （MEDV-13） was used to describe the molecular structures of novel 1,3,5-triazine derivatives. The quantitative structure-activity relationship model （QSAR） was established by leaps-and-bounds regression analysis for the phytocidal activities （PI50）. The correlation coefficients R was 0.888. The QSAR model had both favorable estimation stability and good prediction capability by Rcv^2, FIT, VIF tests. From the three parameters m15, m56, m91 of the model, it could be seen that the molecular structure characteristics, such as -CH2-, 〉CH-, -N- and -X, were the decisive factors affecting the phytocidal activities of the herbicide. The three structure parameters m15, m56, m91 were used as the input neurons of artificial neural network, and the 3：6：1 network architecture was employed. A satisfying model could be constructed with the back-propagation algorithm, and better than MLR-QSAR model with the correlation coefficient R of 0.976. The results show that MEDV-13 has good rationality and efficiency for predicting the phytocidal activities of the herbicide.

关 键 词：分子电性距离矢量 均三氮苯类衍生物 除草活性 定量结构-活性相关 

分 类 号：X131[环境科学与工程—环境科学]