基于高效液相色谱指纹图谱结合化学计量学及机器学习的黑茶产地识别

Identification of dark tea origin based on high performance liquid chromatography fingerprint combined with chemometrics and machine learning

中文关键词:  黑茶  高效液相色谱  化学指纹  产地溯源

英文关键词:dark tea  high performance liquid chromatography  chemical fingerprint  geographical origin traceability

中文摘要:

      目的 建立广西、湖南、四川、陕西和西藏产地黑茶高效液相色谱(high performance liquid chromatography, HPLC)指纹图谱, 并结合化学计量学与机器学习对黑茶进行产地识别研究。方法 采用高效液相色谱-二极管阵列检测器(high performance liquid chromatography-diode array detector, HPLC-DAD)检测48份不同产地黑茶的化学成分, 并建立指纹图谱; 利用没食子酸、表没食子儿茶素、表儿茶素没食子酸酯、表儿茶素、儿茶素、咖啡碱和表没食子儿茶素没食子酸酯7种对照品对图谱共有峰进行指认; 结合化学计量学和不同机器学习算法建立黑茶产地识别模型, 并使用准确率、精确率、召回率及F1分数作为机器学习产地识别模型的评价指标。结果 黑茶指纹图谱共识别出8个共有峰, 指认其中7个成分; 基于指纹图谱8个共有峰峰面积建立的化学计量学和机器学习的产地识别模型中显示, 偏最小二乘法-判别分析模型能识别部分产地黑茶, 并筛选出4个差异标志物, 其预测准确率为54.2%, 逻辑回归(logistic regression, LR)、反向传播神经网络(back propagation neural network, BPNN)、支持向量机(support vector machine, SVM)、随机森林(random forest, RF)和决策树(decision tree, DT)算法模型预测准确率分别为66.7%、90.0%、90.0%、80.0%和80.0%, 其中, SVM模型的产地识别效果最好。结论 我国不同产地黑茶化学成分含量存在一定差异, HPLC指纹图谱结合SVM能够较好对黑茶产地进行溯源研究。

英文摘要:

      Objective To construct the high performance liquid chromatography (HPLC) fingerprints of dark tea in the five major producing areas of Guangxi, Hunan, Sichuan, Shaanxi, and Tibet, and identify the producing areas based on chemometrics and machine learning. Method A high performance liquid chromatography-diode array detector (HPLC-DAD) method was utilized to analyze the chemical components of 48 different origins of dark tea and establish a fingerprint profile. Seven kinds of reference standards including gallic acid, epicatechin, epicatechin gallate, catechin, theaflavin, caffeine, and epigallocatechin gallate were used to identified the peak. A combination of chemometrics and various machine learning algorithms were employed to establish models for the identification of dark tea origins. Accuracy, precision, recall, and F1 score were used as evaluation metrics for the machine learning models. Results The fingerprint profile of dark tea identified a total of 8 common peaks, with 7 components being identified. Based on the peak areas of the 8 common peaks in the fingerprint profile, the chemometrics and machine learning models for the origin identification were established. The partial least squares-discriminant analysis model was able to identify some origins of dark tea and identified 4 differential markers, with a prediction accuracy of 54.2%. The logistic regression (LR), back propagation neural network (BPNN), support vector machine (SVM), random forest (RF), and decision tree (DT) algorithms achieved prediction accuracy of 66.7%, 90.0%, 90.0%, 80.0%, and 80.0%, respectively. The evaluation indicated that the SVM model had the best performance for the origin identification. Conclusion The chemical components of dark tea from different producing areas in China are slightly different, and HPLC fingerprints combined with SVM can better trace the producing area of dark tea.