机器学习算法在不同形态浙贝母与湖北贝母的干法REIMS指纹图谱鉴别分析中的应用研究

Research of machine learning in the application of authenticity discrimination of Fritillariae Thunbergii Bulbus and Fritillariae Hupehensis Bulbus in different form with dry-process REIMS fingerpring

目的:使用快速蒸发离子化质谱(REIMS)指纹图谱与机器学习相关技术对不同形态的浙贝母和湖北贝母进行预测和判别。方法:通过干法灼烧使样品组分形成气溶胶,引入REIMS中,质谱扫描范围m/z 50~1 200,扫描模式为灵敏模式,扫描时间为0.2 s。正离子模式采集,数据记录为continuum模式,测得样品的REIMS指纹图谱数据。通过对数据进行聚类分析、相关性分析、相似度分析、主成分分析,得到数据分布的基本情况,最后建立逻辑回归模型,模型惩罚项参数选择岭回归(l2),优化算法选择拟牛顿法(lbfgs)。结果:测得样品的REIMS指纹图谱具有品种差异的特征性,逻辑回归模型交叉验证和测试集验证准确率均达到1.0,可以准确预测和判别样品的品种。结论:REIMS技术结合机器学习在中药领域的潜在应用前景十分广阔。

Objective:To study and analyze rapid evaporative ionization mass spectrometry(REIMS) fingerprints of samples of Fritillariae Thunbergii Bulbus and Fritillariae Hupehensis Bulbus in different forms for authenticity discrimination with machine learning. Methods: Aerosol formations from the samples by high temperature of dry burning method were ionized and determined by REIMS with m/z 50-1 200 as scanning range in sensitive mode and positive ion mode. The scanning time was 0.2 s and data was recorded as continuous mode. Then the basic situation of REIMS data distribution was studied and analyzed through the methods of cluster analysis, correlation analysis, similarity analysis and principal component analysis. And then logistic regression model with ridge regression(l2) as penalty parameter and quasi-Newton method(lbfgs) as optimization algorithm was established. Results: The REIMS fingerprints of the samples showed the characteristics of variety differences. Both cross validation and test set validation had an accuracy of 1.0, and the logistic regression model could accurately predict and distinguish the varieties of the samples. Conclusion: The application prospect of REIMS technique combined with machine learning in the field of traditional Chinese medicine is very broad.