The studies behind the Atlas
Every dog in the Sniff Atlas comes from somewhere. These are the projects we draw from: CanVAS as the v1 substrate, GRLS as the densest within-breed slice, Darwin's Ark next on the ingest roadmap, plus the catalogue we publish on top of them. And below the cohorts sit the curated knowledge sources, OMIA, VBO, the AVCG, and more, that let the breed and gene pages cite their facts instead of inventing them.
CanVAS
The aggregated canine variant atlas behind every dog in Sniff.
Golden Retriever Lifetime Study
3,000+ Goldens followed for life. The biggest within-breed longitudinal dataset in canine science.
Darwin's Ark
Open-source canine science with 40,000+ owner-uploaded dogs and growing.
NHGRI Dog Genome Project
722 dogs from the Plassais 2019 NIH/NHGRI release — the public-domain breed-reference cohort.
Not cohorts of dogs, but the curated reference layers that let the breed and gene pages cite their facts instead of inventing them: the catalogue of inherited disease, the standard for breed identity, the method for grading what a variant means, the ontologies that connect them across species, and the schema that keeps them interoperable.
OMIA
Online Mendelian Inheritance in Animals: the hand-curated reference for inherited disease, the dog-to-human bridge, and the ground truth our AI pathogenicity is calibrated against. Curated at the University of Sydney since 1995.
VBO
The Vertebrate Breed Ontology: canonical breed identity, alternate names, foundation-stock lineage, and the join that maps disease records onto breeds. Built by the Monarch Initiative with OMIA.
AVCG
The Animal Variant Classification Guidelines: the published, criteria-based framework, modeled on the human ACMG/AMP standard, that decides whether a variant is pathogenic, benign, or uncertain. The rulebook behind every pathogenicity grade we show. Boeykens et al. 2024, Ghent University.
Mondo
The cross-species disease ontology from the Monarch Initiative. It gives a canine disease one stable identity and connects it to its human counterpart, the bridge that lets a dog condition be read against the human disease it models.
uPheno
The unified cross-species phenotype ontology from the Monarch Initiative. It lets a sign in a dog and the same sign in a human be named as one thing, so phenotypes line up across species instead of living in separate vocabularies.
Ensembl
The annotated genome from EMBL-EBI and the Wellcome Sanger Institute, and through Ensembl Compara, the dog-to-human orthology. A clean one-to-one ortholog is the line that connects a canine gene to its human twin; a many-to-many one is where we abstain.
Biolink Model
The shared schema that keeps every entity and relationship interoperable. Invisible on the page, load-bearing underneath it: Biolink is why OMIA, VBO, Mondo, uPheno, and Ensembl coexist as one graph instead of a stack of incompatible tables.
Not curated ontologies, but the genotype and frequency data the population layer is measured against: the reference panel that fills in an array, and the breed-level frequencies that give a carrier rate its denominator.
Dog10K
The international consortium catalogue of canine genetic variation. It is the reference panel the CanVAS genotypes are imputed against, and the baseline for what variation is normal across the dog.
Donner et al. 2023
Over a million dogs screened for 250 disease-associated variants, reported breed by breed. It is the carrier frequency on a breed page, and the sample size that always travels beside it.
gnomAD v4.1
Human gene constraint from 800,000+ people (LOEUF). Read on the human side of the dog-to-human bridge: once a dog gene resolves to a human ortholog, how much that human gene tolerates loss of function. Gene-level, never a dog-variant verdict (INV-57).
ClinVar
Human variant classifications with a review-status confidence rating. Sniff reads only the expert-reviewed (3-star+) pathogenic tier for an affirmative claim, Monarch-modeled, and abstains out loud where the evidence conflicts or is below the bar. Never implies benign (INV-57).