Non-targeted Analysis of Metabolite Differences in Chinese and Foreign Tobacco Extracts based on UHPLC-Orbitrap Mass Spectrometry Combined with Molecular Networking and Machine Learning
Fan Cao, Hua Zhang, Haifeng Shen, Ying Zhu, Yuanqing Ye, Lingling Jiao, Huilin Dong, Yitao Si, Huiyun LiaoIntroduction:
This study aimed to analyze metabolite differences between Chinese and foreign tobacco extracts by establishing a non-targeted identification method integrated with both Compound Discoverer (CD) software and Feature-based Molecular Network (FBMN) datasets.
Methods:
Identification was performed via Ultra-High-Performance Liquid Chromatography (UHPLC)-Orbitrap mass spectrometry. Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA) and Support Vector Machine (SVM) were employed to compare metabolic profiles and explore the feasibility of geographical origin tracing.
Results:
In total, 230 metabolites were annotated, covering sugars, alkaloids, phenolic acids, flavonoids, and others. The OPLS-DA model based on the C-F (CD-FBMN) dataset achieved optimal separation performance (R2 = 0.953, Q2 = 0.729) and screened 25 differential metabolites. The SVM model further annotated 16 characteristic differential metabolites, with 8 markers shared by both algorithms.
Discussion:
The integration of CD and FBMN datasets enabled comprehensive metabolite annotation compared with a single identification strategy. The OPLS-DA and SVM models established with the C-F dataset ensured reliable origin tracing for tobacco extracts.
Conclusion:
This integrated analytical strategy provides a reliable technical reference for the geographical traceability of tobacco extracts and offers valuable insights into the metabolic research of other plants.