Tailored SONAR-MSI: Converting SONAR-MS Data into Pseudoimages for Deep-Learning-Based Natural Products Analysis

Area of Science:

• Analytical Chemistry
• Metabolomics
• Bioinformatics

Background:

• Liquid chromatography-mass spectrometry (LC-MS) is crucial for complex sample analysis.
• Conventional MS data processing is inefficient and loses information, especially for natural products (NPs).

Purpose of the Study:

• To develop a novel workflow, SONAR-MSI, integrating synchronized selected ion acquisition (SONAR), pseudo-mass spectrometry imaging (MSI), and deep learning (DL).
• To improve the quality and efficiency of NP analysis and quality control.

Main Methods:

• Established a SONAR-MSI workflow combining SONAR acquisition with MSI and DL.
• Developed a conversion protocol to transform SONAR-MS data into pseudoimages for convolutional neural networks (CNNs).
• Evaluated data reduction, MS2 quality enhancement, and storage efficiency compared to conventional MS^E.
• Applied a SONAR-MSI-based CNN model for classifying five closely related Ganoderma species.

Main Results:

• SONAR significantly reduces spectral redundancy and improves MS2 quality while minimizing data storage.
• The SONAR-MSI-based CNN model achieved 100% accuracy in classifying Ganoderma species, outperforming feature-table models (91.4%).
• SONAR-MSI enables interpretable pixel-wise mapping of metabolites for visualization and annotation.

Conclusions:

• SONAR-MSI is a robust and scalable approach for DL-assisted metabolomics.
• This workflow enables efficient and information-rich NP analysis and quality control.
• The method offers improved accuracy and interpretability in complex sample analysis.