牛黄类药材REIMS图谱的快速识别研究  

Study of rapid identification of cow-bezoar and its substitutes medicinal herbs usingg REIMS

作　　者：石岩[1] 荆文光 程显隆[1] 魏锋[1] SHI Yan;JING Wen-guang;CCHENG Xian-long;WEI Feng(National Institutes for Food and Drug Control,Beijing 102629,China)

机构地区：[1]中国食品药品检定研究院,北京102629

出　　处：《药物分析杂志》2025年第2期350-360,共11页Chinese Journal of Pharmaceutical Analysis

基　　金：中国食品药品检定研究院关键技术研究基金项目(GJJS-2022-10-2);中国食品药品检定研究院学科带头人培养基金项目(2023X10)。

摘　　要：目的:使用快速蒸发离子化质谱(REIMS)技术对牛黄、培植牛黄、人工牛黄和体外培育牛黄进行测定,并使用机器学习技术对样品的REIMS谱图进行快速识别预测。方法:干法灼烧将样品组分引入REIMS中,质谱扫描范围m/z50~1200;扫描模式为灵敏模式;扫描时间为0.2s。负离子模式采集,数据记录为continuum模式。通过对样品REIMS谱图的数据进行聚类分析和主成分分析,分析样品数据概况。分别建立偏最小二乘判别分析、逻辑回归、决策树、随机森林、自适应提升(分别以逻辑回归和决策树为弱评估器)模型,并通过GaussianCopula、CTGAN、CopulaGAN和TVAE算法仿真合成数据,然后与原训练集数据组成新的训练集用于模型的训练。结果:使用新训练集训练得到的以决策树为弱评估器的自适应提升模型对4种牛黄类药材识别预测能力最好,对测试集识别的准确率为0.97,精确率为0.90,召回率为0.97,F1得分为0.93,ROC面积为1.00。使用模型输出的概率还可以根据药品监管的实际应用场景通过调整概率阈值灵活地使用。结论:使用REIMS技术与机器学习技术联用可以实现牛黄类药材的快速而准确的识别。Objective:To study the rapid identification of cow-bezoar and its substitutes medicinal herbs using the technique of rapid evaporative ionization mass spectrometry(REIMS)couple with machine learning.Methods:The samples were ionized and determined by REIMS with m/z 50-1200 as scanning range in sensitive mode and negative ion mode,0.2 s as scanning time,and using dry burning method.REIMS data of samples was recorded as continuous mode.Then the general situation of REIMS data distribution was studied and analyzed through the methods of cluster analysis and principal component analysis.Some models or algorithms,such as partial least squares discriminant analysis(PLS-DA),logistic regression(LR),decision tree(DT),random forest(RF)and adaptive boosting(AdaBoost,with LR and DT as base estimator respectively)were established.In the models training procedure,simulation synthesis data generated by algorithms of GaussianCopula,CTGAN,CopulaGAN and TVAE joined the original training set data as the new training set.Results:AdaBoost(DT as base estimator)trained with the new training set was the best model which could accurately predict cow-bezoar and its substitutes medicinal herbs.The accuracy for identifying the test set was 0.97,the precision was 0.90,the recall was 0.97,the F1 score was 0.93,and the AUC of ROC was 1.00.The probability output from the model could also be flexibly used by adjusting the probability threshold according to the actual application scenarios of drug regulation.Conclusion:The combination of REIMS technology and machine learning technology can achieve fast and accurate recognition of cow-bezoar and its substitutes medicinal herbs.

关 键 词：牛黄 培植牛黄 人工牛黄 体外培育牛黄 快速蒸发离子化质谱 人工智能 机器学习 真伪鉴别 偏最小二乘判别分析 逻辑回归 决策树 随机森林 自适应提升 仿真数据合成 识别概率 

分 类 号：R917[医药卫生—药物分析学]