MS2MP: A Deep Learning Framework for Metabolic Pathway Prediction from MS/MS-Based Untargeted Metabolomics.

Area of Science:

• Metabolomics
• Bioinformatics
• Computational Biology

Background:

• Untargeted metabolomics generates complex data with low metabolite annotation rates, limiting pathway enrichment analysis.
• Current methods rely on prior metabolite identification, which is often incomplete.

Purpose of the Study:

• To develop a novel deep learning framework, MS2MP, for direct KEGG pathway prediction from tandem mass spectrometry (MS2) spectra.
• To enable pathway enrichment analysis without prior metabolite annotation.

Main Methods:

• MS2MP utilizes a graph neural network architecture to learn relationships between MS2 spectral features and metabolic pathways.
• MS2 spectra are represented as fragmentation tree graphs for analysis.
• The framework was trained on 33,221 experimental MS2 spectra.

Main Results:

• MS2MP achieved high predictive performance (94.1% balanced accuracy in cross-validation, 87.8%-91.2% on independent test sets).
• Achieved exact matches for 97-98% of tested metabolite standards.
• Identified previously overlooked pathway disruptions in transgenic maize, including phenylpropanoid biosynthesis.

Conclusions:

• MS2MP is the first tool for direct metabolic pathway prediction from MS2 spectra.
• This approach enhances pathway enrichment analysis efficiency and biological discovery.
• Facilitates deeper understanding of complex metabolic networks.