结构多样性化合物沸点QSPR模型研究  

QSPR study for compounds with diversity structures

作　　者：周新奇[1] 何勤[1] 赵晨曦[1] 曾映旭[2] 梁逸曾[2] 

机构地区：[1]长沙学院生物工程与环境科学系,湖南长沙410003 [2]中南大学化学化工学院中药现代化研究中心,湖南长沙410083

出　　处：《计算机与应用化学》2008年第6期717-720,共4页Computers and Applied Chemistry

基　　金：长沙市科技计划重点基金(No．K069054-12);湖南省教育厅研究基金(No．06C164);湖南省科技计划基金(No．2007FJ3094)支助．

摘　　要：在定量结构-性质/活性关系(QSPR/QSAR)研究中,分子结构的数值化和建立良好预测的数学模型是2个重要的问题。建立具有良好适应性有实际应用价值的模型是进行QSPR/QSAR的最终目标。本文针对148种来自不同植物挥发油的具有结构多样性化合物,分别采用主成分回归(PCR)、偏最小二乘(PLS)和遗传算法(GA)对其沸点与结构之间的定量结构性能关系进行了研究,用拓扑指数建立了沸点预测模型。结果表明,PLS模型的预测能力最佳,模型训练集的平均相关系数为0.996,平均训练偏差为7.05,检验集的平均相关系数为0.986,平均检验偏差为12.91。It is key that how to convert numerical value from molecular structure and to build good model in quantitative structureproperty/activity relationship （QSPR/QSAR） study. The constructing of QSAR/QSPR prediction model of good compatibility and practical application is our ultimate objective. In the present research, we focus on a set of 148 compounds （ mainly collected from three different volatile oils） to construct predicting models between their structures and normal boiling points by PCR, PLS or GA through topological indices. It showed that the PLS model is the best one in training and prediction. Its training regulations average correlation coefficient is 0. 996, the average training deviation is 7.05. The average testing correlation coefficient is 0. 986 and the average testing deviation is 12. 91.

关 键 词：结构多样性化合物 沸点预测 QSPR/QSAR 拓扑指数 

分 类 号：R284.1[医药卫生—中药学]