Field infrastructure / federation
Federate Sniff into your graph.
Sniff is a governed, cited, cross-species knowledge substrate. Its dog to human bridge, the human clinical and population evidence hung off it, and (when present) the canine disease layer are exported as a Biolink-conformant KGX bundle a steward can ingest. Trust at the seam: every edge carries provenance and evidence level, and nothing that is a prediction is allowed to enter your graph as a fact.
Status: ingestable bundle, not live Translator/Monarch routing.
Registering infores:sniff is human-gated and not done yet.
What this bundle carries
- Dog↔human ortholog spine in this export
19,073 edges · 15,344 high-corroborated (5 methods)
- Human clinical (3★ ClinVar) in this export
558 gene→disease edges through the ortholog
- Human constraint (gnomAD) in this export
16,116 gene nodes carry LOEUF / pLI / missense-Z
- Canine disease bridge (OMIA) in this export
192 OMIA-established disease-gene edges on the ortholog spine
Counts (from the manifest)
- 19,073
biolink:orthologous_to- dog to human ortholog spine, confidence-tiered (15,344 high-corroborated across five methods). - 492
biolink:causes+ 66associated_with_increased_likelihood_of- 3-star ClinVar gene to disease, attached through the resolved ortholog. - 16,116 gene nodes carry gnomAD constraint (LOEUF / pLI / missense-Z) as node properties.
The canine disease bridge
This is the half a cross-species steward cannot get anywhere else. 192 OMIA-established canine disease genes are hung on the
corroborated ortholog spine, so the dog stops being only a recipient of human annotations and becomes a model
organism that contributes evidence. The path is traversable end to end: dog disease to dog gene to corroborated ortholog to human gene to human disease,
every hop cited. Primary source is infores:omia, aggregator infores:sniff; Sniff consumes OMIA, never authors it. Facts only: these are
OMIA's established disease genes, not our hypotheses. The Dog10K natural-model candidates stay barred (INV-54).
Why you can trust it at the seam
- Facts only. Every prediction and candidate atom is excluded by construction. Sniff's Dog10K natural-disease-model candidates are knowledge_level prediction and the exporter bars them, so a hypothesis can never enter your graph as an assertion. Biolink-conformant, 0 conformance issues.
- A provenance quad on every edge. Each edge names its primary_knowledge_source and aggregator_knowledge_source (infores:sniff), its knowledge_level, and its agent_type. An orthology edge also carries the tier and the supporting methods; a ClinVar edge carries its review stars.
- Abstention is recorded, not hidden. Of 29,053 dog genes evaluated, 19,073 were edged and 9,980 abstained with a per-gene reason (8,191 with no ortholog found, the rest one-way / ambiguous / no clean human id). Edged plus abstained equals the input: the exclusion is honest, not lossy.
Sources, each cited
-
infores:ensembl- Ensembl Compara orthology (one2one) - CC-BY-style open, DOI 10.1093/nar/gkae1071 -
infores:orthodb- OrthoDB v11 orthology (concordance rung) - CC-BY-4.0, DOI 10.1093/nar/gkac998 -
infores:oma- OMA orthology (concordance rung) - CC-BY-4.0, DOI 10.1093/nar/gkad1020 -
infores:panther- PANTHER v19.0 orthology (concordance rung, LDO) - CC-BY-4.0, DOI 10.1002/pro.4218; the axis Monarch itself ingests (infores:panther) -
infores:ncbi- NCBI Gene orthologs (concordance rung, synteny) - public domain, DOI 10.1007/s00239-025-10268-2 -
infores:clinvar- ClinVar 3-star gene->disease - public domain, DOI 10.1093/nar/gkx1153 -
infores:gnomad- gnomAD v4.1 gene constraint (node properties) - open, DOI 10.1038/s41586-020-2308-7 -
infores:omia- OMIA Online Mendelian Inheritance in Animals - CC-BY-4.0, DOI 10.25910/2AMR-PV70; canine disease gene associations consumed, never authored by Sniff (INV-47) -
infores:sniff- aggregator (this substrate) - infores id NOT YET REGISTERED (Matt)
For stewards: how to pull and evaluate
- Treat this page as the contract. Counts, facts-only guarantee, reconciliation math, and layer status (live vs pending) are bound to the committed manifest. If a claim is not here, do not assume it is in the bundle.
- Format. KGX TSV (nodes + edges). Biolink-shaped nodes + edges TSVs (KGX). Built by
scripts/build-kgx-export.py. Manifest stamp: 2026-07-20T03:06:37Z . Bundle sha25697f7048fb411f5a7…(recompute over the nodes + edges TSVs you received to confirm match). - How you get the TSV files. The Forge research-compute box is powered down.
The bundle is built on the research git lane and published for stewards at
s3://sniffdog-data/federation/kgx/(sniff_nodes.tsv,sniff_edges.tsv,manifest.json) in the Sniff AWS account, and on the Steward host under~/sniff-research/output/kgx/. This page does not expose a public anonymous HTTPS dump; request access if you do not already have the account. We will not invent a download URL that 404s. - What to check first. (a) edged + abstained = input orthology genes; (b) no prediction/candidate edges; (c) primary sources keep credit; (d) canine disease layer status matches this page (live count or pending). Then sample orthology tiers and a few ClinVar attachments.
- What this is not. Not live Translator routing. Not a score of pathogenicity. Not a claim that dog allele = human allele. Companion instruments for model finding live at Spectra and oncology.
Status: an ingestable bundle, honestly
The KGX TSV bundle (nodes plus edges) is ready to ingest as a facts-only package. It is not yet a live
federation flow: routing Sniff into Translator or Monarch as an aggregator requires registering infores:sniff, which is a deliberate, human-gated step, not an automatic
one. So this is a bundle you can pull and evaluate, with live routing to follow once that registration is done,
on our timeline, never before it is honest to do so.