多氯酚QSAR建模的人工神经网络信息流分析

Flow of information through artificial neural network of chlorophenol QSAR model

针对目前采用人工神经网络 (ANN)建模中存在的问题 ,以多氯酚的定量构效关系 (QSAR)人工神经网络模型研究为基础 ,开展了ANN的信息流分析 .对该网络的连接权值和阈值进行了分析、比较 ,找出了影响网络输出的隐含层的主要节点 ;并由这些隐含层节点进一步确定出主要输入层节点 ;经对网络内部信息传播与分配过程的分析 ,得到ANN模型信息流流径分布图 .

The analysis of information flow through an artificial neural network (ANN) was developed based on the model study of the Quantitative Structure Activity Relationship (QSAR) of chlorophenol. By comparing of the values of connection weights and biases of the ANN model, the key nodes of hidden and input layers for the nodes of output layer were selected, and the distribution figure of information flow through the ANN model was shown. It is useful to discover the more information from ANN model, and further study on the quantitative structure activity relationship and toxicological mechanism of organic chemicals.