中文摘要：

目的研究20种氯苯酚生物毒性的定量构效关系（QSAR）。方法根据分子中原子i的结构特性和键的连接性,在分子图的邻接矩阵基础上构建了一组新的结构信息参数mE（m=0,1）,结合氯苯酚主要电子结构参数,采用主成分分析法对样本数据集进行预处理,将得到的新的样本数据集输入人工神经网络,并构建氯苯酚对花鳞鱼、翻车鱼、鲑鱼、发光细菌等4种水生生物的毒性QSAR模型。结果采用主成分分析法对所有的结构参数（包括结构信息参数mE）进行预处理,减轻了人工神经网络的输入数,并集合了所有参数的信息,同时消除了输入因子的相关性并简化了网络的结构,大大地提高了网络的学习效率和性能。结论主成分分析-人工神经网络的预报精密度优于普通的分析方法。

英文摘要：

Objective To study the quantitative structure-activity relationships（QSAR） on the biology-toxicity of 20 chlorophenols. Methods According to the structural characteristic and the valence connection bonding atom i, a novel structure information index mE （ m = 0,1 ） based on the adjacency matrix of molecule graph was introduced. Combining with electronic structure parameter of chlorophenols,the new principal components analysis （ PCA ） was used as input parameters of ANN analysis to predict the biology-toxicity parameters of chlorophenols, and the QSAR models were established. Results This method was helpful to accelerate the convergence of artifical neural network and simplify the structure of ANN. Conclusion The prediction result was superior to those mothod without using principal component analysis.