人工神经网络和多元回归法预测有机物的生物富集因子  

Estimation of Bioconcentration Factors of Organic Compounds Using Multivariate Regression and Artificial Neural Network

作　　者：冯长君[1] 杨伟华[1] 沐来龙[1] 

机构地区：[1]徐州师范大学化学化工学院,江苏徐州221116

出　　处：《武汉大学学报（理学版）》2007年第6期674-681,共8页Journal of Wuhan University:Natural Science Edition

基　　金：江苏省自然科学基金(05KJD150008);污染控制与资源化研究国家重点实验室基金(PCRRF07009);徐州师范大学培育课题(05PYL04)资助项目

摘　　要：计算了122种非离子性有机物的电性拓扑状态指数En.基于取代基及共轭母体的特征,定义一种新的分子参数——共轭参数B,它们对有机物呈现出良好的结构选择性.将它们与122种非离子性有机物的生物富集因子(lgBCF)拟合,建立令人满意的数学模型:lgBCF=-0.076+1.006B+0.077E38+0.274E9+0.111E39(n′=122,R=0.966,F=408.35,S=0.401).通过Jackknife法检验该模型具有总体稳健性,并能较准确估算与预测有机物的生物富集因子.依此4个结构参数作为神经网络输入层结点,采用4∶10∶1的网络结构,利用BP算法得到一个良好模型,其相关系数R和标准偏差S分别为0.990和0.216,证明该电性拓扑状态指数与共轭参数对于有机物的生物富集因子的预测有效.Based on the characteristics of atom types, Hall＇s electrotopological state indices （EN） were calculated for 122 nonionic organic compounds. On the basis of the characteristics of suhstituents and conjugated matrix, a novel molecular conjugated parameter （B） was defined and calculated for 122 molecules in this paper. En and B show good structural selectivity for organic molecules. A satisfactory relationship between bioconcentration factor （BCF） and E,, B is expressed as. lgBCF= -0. 076＋ 1. 006B＋0. 077E38 ＋ 0. 274E9 ＋0. 111E39（n′=122, R=0. 966, F=408.35, S=0. 401）, which could provide estimation and prediction for the BCF of nonionic organic chemicals. Furthermore, the model is examined to validate overall robustness with Jackknife tests. The four structural parameters are used as the input neurons of artificial neural network, and a 4： 10：1 network architecture is employed. A good model can be constructed with the back-propagation algorithm, the correlation coefficient R and the standard error s are 0. 990 and 0. 216, respectively. This shows that the artificial neural network model for the bioconcentration factor appears to be more reliable than the multivariate regression model. The results show that the new parameter B and electrotopological state index have good rationality and efficiency for the bioconcentration factors of organic compounds.

关 键 词：非离子性有机物 生物富集因子 电性拓扑状态指数 共轭参数 多元线性回归 人工神经网络 定量构效关系 

分 类 号：O625.13[理学—有机化学]