| 利用神经网络法对胺类有机物急性毒性的分类及定量预测 |
| 摘要点击 719 全文点击 1521 投稿时间：1997-06-24 |
| 查看HTML全文
查看全文 查看/发表评论 下载PDF阅读器 |
| 中文关键词 神经网络 构效关系 分子连接性指数 信息理论指数 胺类有机物 急性毒性 预测 |
| 英文关键词 neural networks QSAR molecular connective index information theoretic index amines acute toxicity prediction half lethal dose |
|
| 中文摘要 |
| 运用三层误差反向传播网络对51种胺类有机物进行了结构-毒性关系的研究。选入的结构参数为分子连接性指数(0Xv,1Xv),信息理论指数(RIC,SRIC,RCIC)及分子量等6种均可通过分子拓扑图直接计算获得的指标。毒性参数选用大鼠经口LD50,根据其大小将样本分为3类:高毒、中毒、低毒,在神经网络模型上作出差别归类,并分别对每类进行定量预测。分类预测正确率为90%,定量预测标准差分别为0.083、0.108、0.116.结果表明:神经网络对急性毒性LD50具有良好预测效果,大大优于多元回归分析和判别分析。同时本文还讨论了影响网络性能的一些因素,提出了相应改善措施。 |
| 英文摘要 |
| An application of 3 layers of backpropagation networks to structure-toxicity relationship analysis of 51 kinds of amines has been studied.The structure parameters here are molecular connective indices(0Xv,1Xv),information theoretic indices(RIC,RSIC,RCIC)and molecular weights which all can be calculated directly.Toxicity was indicated by LD50 (rat,P.O.).All those compounds were separated to 3 groups according to LD50:high toxicity、middle toxicity and low toxicity.LD50 of each group was predicted alone.Classification and prediction were carried out using neural networks.The correct rate of classification is 90%,and the standard errors of prediction are 0.083、0.108、0.116 respectively.The results showed that the predicting power of neural networks is strong in the study of QSARs.At the same time,some factors which influence the effect of neural networks were discussed and the ways to improve were given. |
|
|
|
