ARI Disease Metadata Manager

Public editor: https://aurint.ca/ari-editor/

This repository contains the curated autoimmune disease data, ontology artifacts, and generated reports that support the ARI Disease Metadata Manager v2 shown on the public site.

The editor is the main interface for reviewing and maintaining autoimmune disease metadata. It presents the curated registry as a searchable, editable catalog with disease detail panels and source-linked outputs.

What the editor does

• Search diseases, synonyms, and codes
• Create new disease entries
• Search by symptoms
• Browse diseases alphabetically or by tissue target
• Review disease details in a split-pane layout
• Toggle edit mode for record maintenance
• Adjust settings and theme
• Display supporting panels, graphs, and ontology metadata in the footer

Project scope

This repository supports the curated disease model behind the editor:

• Core disease records with ARI IDs and IRIs
• Synonyms, subtypes, definitions, and evidence metadata
• SNOMED, OMOP, DX code, and DOID mappings
• Proposed diseases and proposed changes
• Supplemental per-disease reports and indexes
• Grounding workflows for local ontology matching

Main directories

Path	Purpose	In git?
data/0-archive/	Superseded working material	no
data/1-master/	The curated master list, and a snapshot of the registry site's symptom lists	yes
data/2-databases/	Large raw ontology release dumps — local scratch, re-downloadable	no
data/3-meta-database-sources/	Supporting source tables	yes
data/4-reports/	Generated report outputs	yes
data_model/	Reference spreadsheets and model snapshots	yes
mappings/	Every curated cross-reference judgment	yes
notebook/	Notebook-driven data preparation and grounding scripts	yes
ontologies/	The ARI ontology — the sole authority — as modules (see Ontology modules)	yes
sparql/	SPARQL queries and generated term sets	yes

data/ mixes both kinds, which is worth knowing before going looking for something: 2-databases/ holds roughly 11 GB of downloaded release dumps and is git-ignored in full, while its siblings are curated and tracked. See data/2-databases/README.md.

Key outputs

• data/4-reports/1_Core_ARI_Diseases.xlsx
• data/4-reports/2_Proposed_Diseases.xlsx
• data/4-reports/3_Proposed_Changes.xlsx
• data/4-reports/4_Additional_Info_Index.xlsx
• data/4-reports/5_DOID_Mapping.xlsx
• data/4-reports/6_DOID_Matches_All.xlsx
• data/4-reports/7_SNOMED_Matches_All.xlsx
• data/4-reports/10_Master_List_Comparison.xlsx
• data/4-reports/12_Research_Gaps.xlsx — what each disease still needs, field by field against the data model
• data/4-reports/13_Non_Symptom_Review.tsv — every entry judged not a symptom, for a curator to decide what becomes of it

Grounding workflow

The grounding pipeline in notebook/ari-grounding/ builds local DOID and SNOMED matches from repository data only.

Scripts run in this order:

1. parse_doid_local.py
2. ground_doid.py
3. ground_snomed.py
4. make_match_reports.py

Source data

Primary inputs are the master list in data/1-master/ and the supporting files in data/3-meta-database-sources/, ontologies/, mappings/, and sparql/results/.

Mapping validation

.github/scripts/validate_mappings.py checks the two mapping exports against each other and against the ontology. It needs only the standard library:

python .github/scripts/validate_mappings.py

Add --since main to report only the rows a branch changed. Two workflows run it:

Workflow	Trigger	Scope
Validate mappings	pull requests touching mappings/ or ontologies/	rows the branch added or rewrote — fails the check
Audit mappings	Mondays 07:00 UTC, or on demand	every row, so the standing backlog stays visible
OWL format	pull requests touching ontologies/ or tools/	every ontologies/*.owl is in the canonical layout (see below)

What it catches:

• Structural damage — wrong header, wrong column count, stray tabs, mixed line endings, a byte-order mark, a prefix missing from the SSSOM curie_map.
• Identifiers that are not identifiers — mesh:null and other placeholders, doubled prefixes such as MONDO:MONDO:0014523, ids that do not fit their vocabulary's shape, and ICD-9 codes filed under the icd10cm prefix.
• Contradictions — the same pair recorded as both confirmed and flagged wrong, duplicate rows, one disease under two different labels, dates in the future, unattributed edits.
• Drift between ari.sssom.tsv and ari.equivalencies.tsv, which is how spreadsheet round-trips show up: 362.50 losing its trailing zero, 0111157 losing its leading ones.
• Disagreement with the ontology — a disease id or label that does not match, a cross-reference the curators flagged wrong that is still stored and still served, or a confirmed one that was never stored.
• Curation that disappears without a decision — a synonym, clinical subtype or changelog entry the branch drops rather than adds. A synonym may be retired only by pairing its removal with an ARI_SynonymWithdrawn marker on the same disease, shaped <synonym> | <reason> | <note> where <reason> is subtype, broader, distinct or non-disease; the marker is itself append-only, so the review stays on record.
• Disease ids that are not registry ids. Every ARI_ID has seven digits (ARI:0001234). A disease is renumbered by changing its ARI_ID and adding an ARI_FormerID holding the old one, so the deletion check follows the record to its new id. ARI_FormerID is append-only, and a former id may not still be in use.

The validator reads the ontology as XML, not line by line, so it works on any RDF/XML layout. With --since, an ontology finding counts as new only when the statement on its line (subject, property, value) is new, so re-laying-out the file reports nothing.

Ontology modules

The ontology is a set of files in ontologies/, one per domain. Each opens on its own in Protégé and loads in owlready2 with its imports. Open ari_disease.owl: it imports the rest.

File	Holds	Edited by
ari_core.owl	The key classes and every property declaration	curators (Protégé)
ari_disease.owl	Diseases and their per-disease entries (symptoms, antibodies, antigens, biomarkers, tissues)	the editor app, curators
ari_symptoms.owl	One ARIS_… class per symptom, linked to SYMP and HPO	curators (Protégé)
ari_anatomy.owl	Tissue and system classes, linked to UBERON	curators (Protégé)
ari_biomarkers.owl	Biomarker concepts (empty for now; see notebook/biomarker-concepts/)	curators (Protégé)
ari_symp.owl, ari_uberon.owl	Slices of SYMP and UBERON, generated, never edited by hand	the builder scripts
ari_pr.owl, ari_chebi.owl, ari_hgnc.owl, ari_loinc.owl	Slices of the Protein Ontology, ChEBI, HGNC and LOINC, generated, never edited by hand	build_biomarker_slices.py

A per-disease entry points at its shared concept with sharedConcept. Every property is declared once, in the core. notebook/ontology-modules/README.md covers opening, loading and which tool writes which file; the design is in docs/superpowers/specs/2026-10-18-ontology-modules-design.md.

Canonical OWL layout

The app (owlready2), Protégé and scripts each write RDF/XML in their own layout. So that a diff only ever shows content, every ontologies/*.owl is kept in one layout: the one ROBOT writes, at the version pinned in tools/robot.version. Same content, same bytes, whichever tool saved the file.

tools/format-owl.sh ontologies/ari_disease.owl      # format in place
tools/format-owl.sh --check ontologies/*.owl        # exit 1 if any would change

It needs Java 11 or newer on PATH (Protégé's bundled Java does not count), and it downloads the pinned ROBOT jar into .cache/robot/ on first use. Imports resolve through ontologies/catalog-v001.xml, which Protégé also reads.

Where	What runs
Your clone	The pre-commit hook formats any staged ontologies/*.owl. Install it once: git config core.hooksPath .githooks
The editor app	Formats the file before it pushes a curator's PR branch
OWL format workflow	Pull requests touching ontologies/ or tools/ — fails if a file is not in the canonical layout

If the check fails, run tools/format-owl.sh on the file it names and push again. Scripts that change the ontology edit it as XML and leave the layout to the formatter. Bumping tools/robot.version re-lays-out every file, so do it in its own PR with that reformat commit, and bump the app's deploy/install-robot.sh to the same release.

Working rules

• Prefer the curated master list as the source of truth.
• Keep changes surgical and focused on one data path at a time.
• Use local ontology and report artifacts rather than online sources when generating matching outputs.
• Do not read .env files unless explicitly authorized.

Related documentation

• Data instructions
• Data overview
• Reports overview