Abstract: Based on the structure characteristics of bromine substituents and conjugated matrix, a new path location index ^mP is defined. The electrotopological state indices (E₉) and electronegativity distance vector (M₂₆) of 209 polybrominated diphenyl ethers (PBDEs) are calculated. It shows that ^mP, E₉ and M₂₆ have good structural selectivity for organic pollutant. Then, the multiple regression method is used to establish relationships between the molecular structure of PBDEs and four parameters: gas-chromatographic relative retention time (t_R), n-octanol pair partition coefficient (lgK_OA), supercooled liquid vapour pressures (lgP_L) and toxicity (lgI). With the four predictive models, the correlation coefficients are 0.991, 0.997, 0.998 and 0.904 with the standard deviation of 0.031, 0.104, 0.100 and 0.358, respectively. Three structural parameters are used as the input neurons of the BP neural network and different network architectures to get several satisfying neural network models for predicting t_R, lgK_OA, and lgP_L. The results are better than those by the multiple regression method, which shows significant nonlinear relationships between the physicochemical properties of PBDEs and the three structural parameters.

Key words: polybrominated diphenyl ethers, neural networks, algorithm, model, electrotopological state indices, path location index, physicochemical property

摘要： 基于溴取代基及共轭母体的结构特征，定义一种新的路径定位指数^mP，并计算了209种多溴联苯醚分子的电性拓扑状态指数E₉和电性距离矢量M₂₆，这3个指数对有机污染物呈现出良好的结构选择性。采用多元回归方法对多溴联苯醚（PBDEs）的色谱保留时间（t_R）、正辛醇/空气分配系数（lgK_OA）、超冷流体蒸气压（lgP_L）及其毒性等性质与分子结构指数进行关联，建立的4个预测模型的相关系数分别达到0.991、0.997、0.998和0.904，标准偏差分别为0.031、0.104、0.100和0.358。将3个指数作为BP神经网络的输入层神经元，采用不同的网络体系结构，获得了令人满意的预测理化性质的神经网络模型，所得结果明显优于多元回归方法，结果显示多溴联苯醚的理化性质与⁰P、E₉和M₂₆具有良好的非线性关系。

关键词: 多溴联苯醚, 神经网络, 算法, 模型, 电性拓扑状态指数, 路径定位指数, 理化性质