Abstract: A series of polyhalogenated biphenyls have been used to develop quantitative structure-retention relationship for their gas and liquid chromatographic retention index by using two 2D descriptors of the atom type electrotopogical state index and the molecular electronegativity distance vector based on 13 atomic types. QSRR of 55 kinds of polyhalogenated biphenyls models were built by multiple liner regression and artificial neural network. The results show that using artificial neural network method is better than [KG*4]using [KG*4]multivariate [KG*4]linear [KG*4]regression, the [KG*4]predictive [KG*4]correlation [KG*4]coefficient R can reach above 0.99. It is demonstrated that using artificial neural network method can accurately predict polyhalogenated biphenyls gas and liquid chromatographic retention index.

Key words: Quantitative structure-retention relationship, Polyhalogenated biphenyls, Multiple liner regression, Artificial neural network

摘要： 将卤代联苯化合物作为研究体系，利用基于原子类型的电子拓扑结构（E-state）和基于13种原子类型的电性距离矢量描述子（MEDV-13）作为描述符，分别应用多元线性回归、人工神经网络中的误差反向传播神经网络和径向基函数神经网络的方法建立了55种卤代联苯化合物的QSRR模型。使用人工神经网络的方法预测的结果比多元线性回归的方法的结果稍好，相关系数R可以达到0.99以上，说明使用人工神经网络的方法能够准确地预测卤代联苯化合物的气相色谱和液相色谱的保留指数。

关键词: QSRR , 卤代联苯化合物 , 多元线性回归 , 人工神经网络