Analysis of Primary Liquid Chromatography Mass Spectrometry Data by Neural Networks for Plant Samples Classification

Area of Science:

• Metabolomics
• Chemotaxonomy
• Analytical Chemistry

Background:

• Plant-derived secondary metabolites are crucial in traditional medicine and research.
• Adulteration of herbal drugs with different plant parts is a significant challenge.
• Liquid chromatography and mass spectrometry offer powerful tools for plant sample analysis.

Purpose of the Study:

• To develop and evaluate novel fingerprint analysis approaches for distinguishing plant parts within the Apiaceae family.
• To identify specific chemical markers characteristic of different plant organs (roots, stems, leaves, fruits).
• To address data limitations in metabolomic studies through data augmentation.

Main Methods:

• Utilized liquid chromatography-low-resolution mass spectrometry for plant sample fingerprinting.
• Applied two distinct analytical approaches: Support Vector Machine (SVM) and Neural Networks (NNs).
• Implemented minimal data preprocessing and explored five data augmentation techniques.

Main Results:

• Achieved comparable F1-scores around 0.75 using both SVM and NN methods.
• Successfully identified eight marker compounds (chlorophylls, lipids, coumarin apio-glucosides) associated with specific plant parts.
• Demonstrated the effectiveness of the proposed methods in preserving information content.

Conclusions:

• The developed fingerprint analysis approaches are simple, information-saving, and effective for plant part classification.
• These methods can be broadly applied to various metabolomic tasks, including quality control and authentication of herbal materials.
• The identification of marker compounds enhances the understanding of plant organ-specific metabolomic profiles.