Knowledge and data-driven two-layer networking for accurate metabolite annotation in untargeted metabolomics

Area of Science:

• Metabolomics
• Bioinformatics
• Systems Biology

Background:

• Metabolite annotation is difficult due to diverse structures, especially for compounds without chemical standards.
• Network-based methods offer powerful solutions for metabolite identification in untargeted metabolomics.

Purpose of the Study:

• To develop an advanced network-based strategy for improved metabolite annotation in untargeted metabolomics.
• To enhance the coverage, accuracy, and computational efficiency of metabolite identification.

Main Methods:

• Developed a two-layer interactive networking topology integrating data-driven and knowledge-driven networks.
• Curated a comprehensive metabolic reaction network using graph neural network-based predictions.
• Integrated experimental data via MS1 matching, reaction mapping, and MS2 similarity constraints.

Main Results:

• Achieved over 10-fold improvement in computational efficiency for annotation propagation.
• Annotated over 1600 seed metabolites and >12,000 putatively annotated metabolites in biological samples.
• Discovered two previously uncharacterized endogenous metabolites not present in existing databases.

Conclusions:

• The developed strategy significantly enhances metabolite annotation coverage, accuracy, and efficiency.
• MetDNA3 provides a robust and efficient tool for metabolite identification in complex biological systems.
• This approach facilitates the discovery of novel metabolites and advances metabolomic research.