MS-Net: Multi-Similarity-Based Network Annotation for Untargeted Metabolomics.

Area of Science:

• Metabolomics
• Bioinformatics
• Computational Chemistry

Background:

• Metabolite annotation is a major challenge in untargeted LC-MS metabolomics, with limited experimental spectral libraries.
• In silico tools often produce high false annotation rates due to inaccurate top-ranked predictions.

Purpose of the Study:

• To develop an accessible workflow, MS-Net, for prioritizing metabolite annotations in large candidate spaces.
• To integrate mass spectral similarity, molecular structure similarity, and taxonomic knowledge for improved annotation accuracy.

Main Methods:

• Developed MS-Net, a workflow integrating mass spectral similarity networks, Tanimoto structural similarity, and taxonomic filtering.
• Utilized a composite Link Score combining structural, spectral, and computational evidence for annotation prioritization.
• Applied the workflow to Cannabis sativa extracts for compound identification and pathway reconstruction.

Main Results:

• MS-Net assigned 1275 compounds from over 118,000 initial candidates in Cannabis sativa extracts.
• 53% of annotations were rescued from ranks 2-50, correcting initial in silico ranking deficiencies.
• Successfully reconstructed known cannabinoid biosynthetic pathways, confirming biological coherence of annotations.

Conclusions:

• MS-Net enhances metabolite annotation confidence and accuracy in untargeted LC-MS metabolomics.
• The workflow effectively prioritizes annotations by integrating diverse similarity metrics and taxonomic information.
• MS-Net is available as a reproducible, offline KNIME workflow for systems biology applications.