目的 建立蒲黄炭炮制程度判别和多成分定量的近红外(Near infrared，NIR)快速分析方法。方法 制备"不及""适中""太过"3种炮制规格的蒲黄炭样品186批，采集不同规格蒲黄炭的NIR光谱，分别采用偏最小二乘判别分析(Partial least squares discriminant analysis，PLS-DA)、K-最近邻(k-nearest neighbor，kNN)和支持向量机(Support vector machine，SVM)建立蒲黄炭炮制程度的判别分析模型，以准确度(Accuracy，ACC)和错误率(Error rate，ER)对不同类型判别模型的判别效果进行评价。然后采用偏最小二乘(Partial least squares，PLS)建立蒲黄炭中原儿茶酸、3-羟基苯甲酸、4-羟基苯甲酸、壬二酸、槲皮素、槲皮素-3-O-(2G-α-L-鼠李糖基)-芸香糖苷、山柰酚、山柰酚-3-O-(2G-α-L-鼠李糖基)-芸香糖苷、异鼠李素、香蒲新苷、异鼠李素-3-O-新橙皮糖苷和柚皮素的定量分析模型，计算校正集和验证集相关系数(real、rpre)、校正集误差均方根(Root mean square error of calibration，RMSEC)、验证集误差均方根(Root mean square error of prediction，RMSEP)和性能偏差比(The ratio of prediction to deviation，RPD)对 PLS 模型预测性能进行评价。结果 与PLS-DA和KNN模型相比，SVM模型对不同炮制程度蒲黄炭判别效果最优，校正集和验证集的ACC分别为90。08％和93。44％，ER分别为9。08％和5。21％。蒲黄炭中12种化学成分的PLS模型rcal和rpre均大于0。9，RPD均大于2。3。结论 本文采用NIR光谱并结合化学计量学建立蒲黄炭炮制程度判别分析和多成分定量分析的快速检测方法，该方法快速、无损、准确，为快速判断蒲黄炭炮制程度和分析化学成分变化，保证蒲黄炭炮制工艺的稳定性和饮片质量的可控性提供科学依据和方法支撑。
Rapid Discriminate Analysis and Quantitative Analysis Methods for Carbonized Typhae Pollen Using Near Infrared Spectroscopy Coupled with Chemometrics
Objective To establish the rapid discriminate analysis and quantitative analysis methods for carbonized Typhae Pollen(CTP)using near infrared spectroscopy coupled with chemometrics.Methods A total of 186 batches of CTP samples were prepared and categorized into three group including light degree CTP,moderate CTP,and heavy degree CTP.The NIR spectra of these samples were characterized.Then partial least squares-discriminant analysis(PLS-DA),k-nearest neighbors(KNN),support vector machine(SVM)algorithms were separately applied to build the discriminant models.The performance of discriminant models was evaluated in terms of the accuracy(ACC)and the error rate(ER).The partial least squares algorithm was used to establish the quantitative model for the prediction of 3,4-dihydroxybenzoic acid,3-hydroxybenzoic acid,4-hydroxybenzoic acid,azelaic acid,quercetin,quercetin-3-(2G-rhamnosylrutinoside),kaempferol,kaempferol-3-(2G-rhamnosylrutinoside),isorhamnetin,typhaneoside,isorhamnetin-3-O-neohesperidin,and naringenin of CTP samples.The correlation coefficients(rcal,rpre),the root mean square error of calibration(RMSEC),the root mean square error of prediction(RMSEP),and the ratio of performance deviation(RPD)were calculated to assess the PLS model.Results Compared with PLS-DA and KNN models,the SVM model yielded the best classification.The ACC of SVM model was 90.08％for calibration set and 93.44％for prediction set,while the ER was 9.08％for calibration set and 5.31 for prediction set.The values of rcal and r for PLS models were greater than 0.9,and the RPD of that was greater than 2.3.Conclusion In this study,the NIR spectroscopy coupled with chemometrics was firstly applied to develop the rapid discriminant analysis and quantitative analysis methods for CTP.The NIR-based method is rapid,non-destructive,and accurate,and it provides the scientific basis and method support for the rapid judgment of the processing degree of CTP and analysis of the changes of chemical components to ensure the quality of CTP.
万方数据
