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

Confident metabolite annotation remains a critical bottleneck in untargeted LC–MS metabolomics, with experimental spectral libraries covering only 5–20% of detected features. While in silico tools generate extensive candidate lists per feature, top-ranked predictions frequently fail to reflect true molecular identities, leading to high false annotation rates. We present multi-similarity Network-based annotation (MS-Net), an accessible workflow that integrates mass spectral similarity networks, molecular structure similarity (Tanimoto metrics), and taxonomic knowledge to prioritize annotations within vast candidate spaces. High-confidence annotations from authentic standards, spectral libraries, and taxonomically filtered candidates seed iterative propagation throughout mass spectral similarity networks. The workflow employs a composite Link Score combining structural, spectral, and computational evidence to rescue correct annotations from lower-ranked positions. Applied to Cannabis sativa extracts (2595 features to 1297 after filtering), MS-Net assigned 1275 compounds from an initial candidate space of over 118,000 structures. Notably, 53% of final annotations were rescued from ranks 2–50, demonstrating correction of initial in silico ranking. The workflow successfully reconstructed known cannabinoid biosynthetic pathways, validating biological coherence. MS-Net is freely available as a KNIME workflow with complete documentation at https://forge.inrae.fr/metatoul/equipe-agromix/ms-net, enabling reproducible, offline annotation suitable for systems biology integration.