Spectral networking for dark matter annotation
Map structural relationships between named and unnamed features across your samples. Cosine similarity and neutral-loss propagation turn isolated annotations into a connected molecular map.
Shared structure
Features linked by structural motifs, adduct variants, or common fragmentation pathways form a connected map, not isolated points.
Inherited context
When one node earns an MSI Level 1a identification, its neighbours inherit structural context automatically.
Named dark matter
Unknown features become "dark matter" with a family name, narrowing the search space for subsequent annotation.
Cosine Similarity
The MS² spectrum of each feature is compared pairwise across the dataset. A cosine score ≥ 0.85 between two spectra indicates they share a common substructure. VeroMass computes similarity in both ESI+ and ESI− independently, capturing complementary views of the same molecule.
Molecular Families
Connected components in the similarity graph form molecular families: related glycosides, acylated derivatives, or oxidation products that differ by a known neutral loss. A single Level 1a assignment propagates structural knowledge to every member of the family.
Neutral Loss Propagation
Network edges are annotated with the mass difference between connected nodes. Neutral losses of 162.053 Da (hexose), 146.058 Da (deoxyhexose), and 176.032 Da (glucuronide) are automatically flagged — turning a collision cell into a structural classifier.
Propagation rules
Cosine ≥ 0.85 + consistent mass shift ≤ 5 ppm + RT correlation → structural family assignment. In any untargeted metabolomics experiment, the majority of detected features have no match in any reference library. When an unknown feature shares high cosine similarity with a Level 1a node and exhibits a consistent mass shift, it inherits the structural class of its neighbour — moving from "unknown" to "likely flavonol diglycoside" or "probable hydroxycinnamate ester," a hypothesis that can be tested. Each propagated annotation carries a confidence score derived from the distance to the anchor node and the number of independent edges supporting the connection.
Force-Directed Layout
Nodes repel and edges attract — the graph automatically organises by structural similarity. Move, pin, and zoom any node. Layout parameters are adjustable for dense or sparse datasets.
Multi-Layer Views
Toggle between ESI+ and ESI− networks, overlay retention time gradients, colour by compound class, and filter by cosine threshold — all without regenerating the graph.
Selection to Evidence
Click any node to pull up its MS¹ and MS² spectra, retention time, adduct series, and gate status. Click an edge to see the cosine alignment between the two spectra side by side.
The network graph, annotations, and edge tables are exportable in formats that feed directly into downstream analysis:
- Export to Cytoscape / Gephi as GraphML or CSV edge tables
- Incorporate family assignments into Python or R workflows
- GNPS-compatible export for cross-platform validation
- Public repository deposition (MassIVE)
- One interface across GraphML, JSON, GNPS & CSV