环境化合物对鱼类毒性的定量构效关系研究  

QSAR study of toxicity on fish based on molecular descriptors

作　　者：王妲[1] 李燕[2] 丁君[1] 王媛[1] 王永华[1] 常亚青[1] 

机构地区：[1]大连水产学院,农业部海洋水产增养殖学与生物技术重点开放实验室,辽宁大连116023 [2]大连理工大学,材料化工系,辽宁大连116024

出　　处：《计算机与应用化学》2009年第4期473-478,共6页Computers and Applied Chemistry

基　　金：辽宁省教育厅资助项目(2008149)

摘　　要：本研究基于定量构效关系方法预测环境化合物对鱼类的毒害（50％ Lethal Concentration，LC50），并确定影响毒性关键分子的结构特征及几种模拟方法的比较。构建114个化学分子的数据集，随机选取85个75％分子为训练集，剩下的29个分子作为检验集，每个化学分子计算194个分子参数，分别采用逐步多元线性回归分析法（Principal Component Regression, PCA）、主成分回归法（Principal Component Regression, PCA）和偏最小二乘法（Partial Least Square，PLS）构建定量结构一毒害关系（Quantitative structure-activity relationships，QSTR）模型。用逐步多元线性回归分析法得出的训练集和预测集的实验值-logLC50与预测值-logLC50的相关系数分别为Rtr^2=0．86，Rte^2=0．83，说明该模型可靠性和鲁棒性较高；主成分回归法用8个主成分，其训练集和预测集的实验-logLC50与预测-logLC50的Rtr^2=0．81，Rte^2=0．77；偏最小二乘法用了5个潜变量，其训练集和预测集的实验-logLC50与预测-logLC50的Rtr^2=0.88，Rte^2=0.85。MLR方法得出化合物对鱼类的毒害影响较大的分子参数，主要分属电拓扑状态参数（SssO，SsCl，SdCH2，SsNH2）、分子连接指数（Xv0）以及修正Kappa指数（Ka2）。以上研究对预测环境化合物的鱼类毒害（LC50），以及从机理上加深对有机物的毒性作用机理提供重要价值。The aim of this work is to predict the toxicity （LC50） of the environmental compounds on fish based on QSAR methods, and to determine the key descriptors influencing the toxicity, and also to compare several mathematical methods in building models. The QSAR models were built based on a dataset of 114 diversified compounds. We randomly selected 75% （85 compounds） as training set and the rest 25% （29 compounds） as the test set. For each molecule, 194 molecular indices were calculated. The models were developed using multiple linear regression （ MLR）, principal component analysis （PCA） and partial least squares （ PLS）, respectively. For the MLR model, the correlation coefficients R^2 between the experimental and predicted -logLC50 values of training set and test set are Rtr^2 = 0. 86 and Rte^2 = 0.83, respectively, explaining that this model has the high reliability and robustness. For the PCA model, using 8 PCs, the correlation coefficients R^2 between the experimental and predicted -logLC50 values of training set and test set are Rtr^2 = 0. 81 and Rte^2 = 0.77; for the PLS model, using 5 latent components, the correlation coefficients R^2 between the experimental and predictedlogLC50 values of training set and test set are Rtr^2 = 0. 88 and Rte^2 = 0. 85. The key molecular structure parameters extracted by MLR which influence the toxicity on fish are electro-topological state indices （ SssO, SsCl, SdCH2, SsNH2 ） , molecular connectivity （ Xv0 ） and Kappa index （Ka2 ）. The present work should be valuable for evaluating the toxicity of environmental compounds on fish, as well as to deepen our insights to the toxic mechanism of compounds.

关 键 词：水毒理学 定量结构-毒性关系 多元线性回归分析 主成分分析 偏最小二乘法 

分 类 号：O6[理学—化学] TP393[自动化与计算机技术—计算机应用技术]