分子距边矢量结合神经网络法预测二(口恶)(口英)类化合物PCDFs的logK_(ow)值

Predicting the LogKow Using Molecular Distance - edge Vector Combined with Artificial Neural Network (ANN) Method

采用以基团间距离、边数为基准的分子距边矢量作为二(口恶)(口英)类化合物多氯代二苯并呋喃(PCDFs)的分子结构描述符,结合反向传播人工神经网络方法建立了PCDFs的正辛醇/水分配系数(logKow)与分子结构描述符之间的定量关系模型。结果表明,所采用的结构描述符对于分子结构具有很好的区分能力,所建立的模型对于校验样本的预测精度较高。利用所得模型,对文献尚未报道logKow实验值的其他所有85种PCDFs给出了预测值。

Polychlorinated dibenzofurans (PCDFs) are highly concerned as persistent organic pollutants and suspected endocrine disrupters, whose property logKow plays an important role for its environmental risk assessment. In this paper, a novel molecular distance - edge vector (VMDE, μ in short) was introduced as structure descriptors, then a quantitative relationship of high accuracy was established with back - propagation artificial neural network method. An additional sample set was used to test the model with quite good results. With the established model, the logKow values of the other 85 PCDFs not belonging to the modeling sample set were given, these values has not been reported in literatures.