关键词: 人工神经网络, 分子结构描述符, 吡喃酮类化合物, 定量结构活性相关

Abstract: MOPAC-PM3 algorithm in Chemoffice 2004 was used to calculate quantum chemical structure parameters of pyranones，and the quantization parameter selection were used as descriptors of pyranones. With molecular descriptors，the structure of pyranones compounds were characterized and anti human immunodeficiency virus（HIV）activity was predicted. The model on molecular descriptors and biological activity was established with RBF neural network of artificial neural network. When sp= 0. 41，the results showed that the predicting variance of MSE for network training set is nearly 0，and the network simulation and prediction of MSE was 0. 006 6，the total MSE was 0. 000 7. The results showed that the RBF neural network had the feature of high digital approximation，which improved the predition accuracy of structure of pyranones compounds.

Key words: artificial neural network, molecular descriptors, pyranones compounds, quantitative structure-activity relationship