基于色谱指纹图谱和化学计量学建立党参的定性模型  

Developing the classification model of radix Codonopsis pilosula according to cultivation area based on chromatographic fingerprint and chemometrics

作　　者：刘兰萍[1] 李波霞[1] 魏玉辉[1] 段好刚[1] 武新安[1] 

机构地区：[1]兰州大学第一医院药剂科,甘肃兰州730000

出　　处：《兰州大学学报（医学版）》2012年第3期38-41,共4页Journal of Lanzhou University（Medical Sciences）

基　　金：甘肃省中医药管理局资助项目(GZK-2008-32)

摘　　要：目的建立来源于不同产地党参的分类模型。方法采用近红外光谱技术结合化学计量学方法鉴别党参产地,采用主成分分析考察分类的可能性和趋势,采用随机森林算法建立分类模型并选择出特征变量,再用所选变量建立径向基函数神经网络模型。结果对于随机森林算法,袋外估计的准确率为86.96%,检验集准确率为100%;基于随机森林算法选出的特征变量,径向基函数神经网络模型的训练集和检验集的准确率都高达100%。结论初步研究表明,色谱指纹图谱结合化学计量学能有效的区分不同产地的党参。Objective To establish a method for the classification of Codonopsis pilosula from three cultivated places.Methods The origins of Codonopsis pilosula were identified by the use of high-performance liquid chromatography coupled with chemometrics.Principal component analysis was applied to investigate the feasibility and tendency of classification.Random forests was used to build the classification model and to select the most effective variables,which was used to build radial basis function neural networks model（RBF）.Results For random forests classifier,an overall out-of-bag estimation of the correct classification rate of 86.96%and 100% were obtained.RBF showed the accuracy rate were as high as 100%for the training as well as the test data set.Conclusion The results indicated that the chromatographic fingerprint combined with chemometrics can efficiently discriminate Codonopsis pilosula from various cultivated places.

关 键 词：党参 指纹图谱 随机森林 径向基函数神经网络 

分 类 号：R282.5[医药卫生—中药学]