MetabolomicsHub skill

[lfnothias]

Adding a skill for https://www.metabolomicshub.org/ cross-repositories crawling
https://metabolomicshub.github.io/mhd-model/
https://www.metabolomicshub.org/api/submission/summary
https://github.com/MetabolomicsHub

[lfnothias]

🤖 Curator triage → staged for wave community-2026-06 (deferred)

Field	Resolved
Identity	MetabolomicsHub (MHD) — common data model + MetaboLights/GNPS/Metabolomics Workbench integration
DOI	❌ none found (no CITATION.cff / .zenodo.json on mhd-model)
Open access	✅ code is Apache-2.0
Category	Formats / data infrastructure
Decision	defer-pending-info

This is data infrastructure, not a paper or a versioned software release — even under the new software-DOI policy there's no citable identifier to attach. Two paths: (a) request a citable DOI from upstream, or (b) route it to an ASB skill/tool pack rather than the paper corpus. Staged with a note at collections/metabolomics/proposals/wave-2026-06-community.yaml.

[lfnothias]

Staged as metabolomicshub-cross-repository-dataset-search in #39, held for review — it sits in proposals/skills/ with status: hold and is not part of the published collection until a maintainer promotes it.

The skill covers finding studies across MetaboLights, Metabolomics Workbench and GNPS/MassIVE in one query, then filtering to what a reanalysis can actually use: selection criteria written down before the search rather than after, vendor-only raw data treated as a study-level exclusion, cross-repository duplicates resolved on title and submitter, and accessions recorded rather than counts. It states that hub coverage is bounded by what the repositories announce, so a negative result is not evidence of absence, and that normalised instrument strings should be confirmed against each study's own record.

Grounded on MetabolomicsHub/mhd-model rather than on a paper, so it carries the repository provenance tier added for exactly this case. Licence read from the repository: Apache-2.0, open tier.

Review welcome on the procedure and the limitations before promotion — you know the hub's coverage better than the corpus does.

[lfnothias]

Rewritten against the live service in #40, after checking it rather than the draft's assumptions.

The URL this skill named, https://www.metabolomicshub.org/api/, returns 404. The service is at /api/submission/v0_1, indexes 6,796 datasets, and does expose search — a clause DSL over 40 fields, which the draft did not describe. Three specific corrections:

• Polarity is filterable, but as a parameter_pair with type_name: "scan polarity", not a top-level field.
• File format is not indexed. The draft listed "filter on open raw formats" as a query step; it has to be done client-side from the files.extensions block each result carries.
• Coverage is repository-dependent and the exclusions are silent. GNPS/MassIVE (500 datasets) carries no structured instrument or parameter metadata, so any instrument or polarity filter drops it whole. facet_omics_types is MetaboLights-only: filtering on it returned 2,331 datasets, and adding dataset_repository = metabolights returned the same 2,331.

The skill now carries the coverage table and a procedure step that measures each filter's per-repository exclusion. A companion skill covers the depositor's side (metabolomicshub-announcement-authoring), which is where that uneven coverage originates.

Observations are dated against index v0.0.1 and recorded in verified_against, since both the index and the field list move. Still status: hold.