HolobiomicsLab spectral-library-matching-and-ranking
Summary
This skill applies Spec2Vec embeddings to compute cosine similarity between Mass2Motifs and a spectral library, retrieving and ranking top-matched annotated spectra or reference entries to generate per-motif annotation records with confidence metrics.
When to use
When you have discovered Mass2Motifs from MS2LDA topic modeling and need to automatically annotate them by finding the most structurally similar known spectra in a reference library. Use this skill when manual annotation is infeasible at scale and you want reproducible, quantified similarity-based substructure suggestions.
When NOT to use
- The motifset has not yet been trained or does not exist in JSON format — first complete MS2LDA LDA modeling.
- You are working with private, unpublished spectra not represented in the Spec2Vec-indexed library — library coverage bias will degrade annotation quality.
- You require manual expert curation rather than automated ranking — use this skill to accelerate triage, not to replace human validation.
Inputs
- trained motifset JSON file (motifset_optimized.json)
- pre-trained Spec2Vec model from Zenodo
- Spec2Vec embeddings from Zenodo
- Spec2Vec-indexed spectral library database
Outputs
- per-motif annotation records in JSON format
- motif-to-label mappings with similarity scores
- confidence metrics per annotation
- MS2LDAViz-compatible annotation output
How to apply
Load the pre-trained Spec2Vec model and embeddings from the designated Zenodo repository (zenodo.org/records/15688609), then load the trained motifset JSON file (motifset_optimized.json) containing all discovered Mass2Motifs. For each Mass2Motif, compute Spec2Vec similarity scores against the Spec2Vec-indexed spectral library using the loaded model. Retrieve the top-ranked annotated spectra or reference entries for each motif based on a cosine similarity threshold. Generate annotation records mapping each motif identifier to suggested labels, similarity scores, and confidence metrics. Export the per-motif annotation output in JSON format compatible with MS2LDAViz for downstream visualization and validation.
Related tools
- Spec2Vec (computes cosine similarity scores between Mass2Motifs and reference spectra embeddings)
- MAG (Mass2Motif Annotation Guidance) (orchestrates automated annotation of Mass2Motifs using Spec2Vec similarity ranking) — https://github.com/vdhooftcompmet/MS2LDA
- MS2LDA (generates the trained motifset that serves as input to this annotation skill) — https://github.com/vdhooftcompmet/MS2LDA
- MotifDB (searchable database of reference motifs and annotations used for lookup and comparison)
- Python (host language for loading models, computing similarities, and exporting JSON annotations)
Evaluation signals
- Verify that output JSON contains all motif IDs from the input motifset with no missing records.
- Check that each annotation record includes a similarity score field with numeric values in the range [0, 1] (cosine similarity bounds).
- Confirm that top-ranked matches correspond to known substructures consistent with typical fragmentation patterns for the compound class.
- Validate that JSON output conforms to MS2LDAViz schema and loads without parsing errors in the visualization interface.
- Cross-reference a sample of high-confidence matches (e.g., similarity > 0.8) against the MotifDB or literature to assess biological plausibility.
Limitations
- Annotation quality depends entirely on the coverage and accuracy of the pre-trained Spec2Vec model and indexed library — novel substructures absent from the training set will receive poor or incorrect matches.
- Cosine similarity thresholds are not automatically optimized and must be tuned empirically; no principled default is stated in the article.
- The skill assumes the motifset JSON structure and field names match the expected format; incompatible or corrupted input files will cause silent failures or malformed output.
Evidence
- [other] For each Mass2Motif, compute Spec2Vec similarity scores against the Spec2Vec-indexed spectral library using the loaded model.: "For each Mass2Motif, compute Spec2Vec similarity scores against the Spec2Vec-indexed spectral library using the loaded model."
- [other] Download and load the pre-trained Spec2Vec model and embeddings from the Zenodo repository (zenodo.org/records/15688609).: "Download and load the pre-trained Spec2Vec model and embeddings from the Zenodo repository (zenodo.org/records/15688609)."
- [other] Load the trained motifset JSON file (motifset_optimized.json) containing all discovered Mass2Motifs.: "Load the trained motifset JSON file (motifset_optimized.json) containing all discovered Mass2Motifs."
- [other] Retrieve the top-ranked annotated spectra or reference entries for each motif based on cosine similarity threshold.: "Retrieve the top-ranked annotated spectra or reference entries for each motif based on cosine similarity threshold."
- [other] Generate annotation records mapping each motif identifier to suggested labels, similarity scores, and confidence metrics.: "Generate annotation records mapping each motif identifier to suggested labels, similarity scores, and confidence metrics."
- [methods] Automated Mass2Motif Annotation Guidance (MAG) with Spec2Vec: "Automated Mass2Motif Annotation Guidance (MAG) with Spec2Vec"
- [other] MS2LDA provides automated annotation of Mass2Motifs using MAG with Spec2Vec, which loads a trained motifset together with the Spec2Vec model from a Zenodo repository to output per-motif annotation records.: "MS2LDA provides automated annotation of Mass2Motifs using MAG with Spec2Vec, which loads a trained motifset together with the Spec2Vec model from a Zenodo repository to output per-motif annotation"