The honest knowledge substrate for canine genetics, and its bridge to human medicine.
Open, cited, and queryable. Every claim traces to its source, or it says it does not know.
The problem it solves
Animal genetics has needed one thing for years: a central, cited, provenance-safe reference layer that connects genes, variants, diseases, breeds, and their human analogs, that anyone can query and actually trust. Until now that knowledge has lived scattered across papers, databases, and spreadsheets, with no honest way to ask it a question and know where the answer came from.
Sniff is that layer. It is built on the field's own canonical sources, it holds a single discipline (cite the science or say you do not know), and it is instantiated on the dog first, with the architecture to extend across species.
What it is built on
Sniff renders and cites the field's established, open resources. It does not score variants with a model of its own. That is the whole trust model. You are never asked to believe Sniff, you are shown exactly whose work each answer stands on.
- OMIA , the documented disease and variant catalogue, used under CC-BY.
- Vertebrate Breed Ontology (VBO) , canonical breed identity and lineage.
- Mondo , the cross-species disease ontology (Monarch Initiative), and the bridge to human disease.
- uPheno , the unified cross-species phenotype ontology (Monarch).
- AVCG (Boeykens et al. 2024) , the animal variant-classification guidelines, the field's equivalent of the ACMG/AMP standard.
- Biolink Model , the schema that keeps every entity and relationship interoperable.
- Ensembl , gene models and dog-to-human orthology (Compara).
- gnomAD v4.1 , human gene constraint (how much the human counterpart tolerates loss of function), read off the resolved ortholog.
- ClinVar , expert-reviewed human variant classifications (Monarch-modeled), abstained out loud where the evidence is unsettled.
- CanVAS , the harmonized, imputed canine variant dataset behind the population map.
- Donner et al. 2023 , breed-level carrier frequencies with sample size on every value.
- Dog10K , the open canine genome reference panel.
- Darwin's Ark (Morrill et al. 2022) , an open research genome release.
- NHGRI (Plassais et al. 2019) , an open research genome release.
The population layer
Most of the sources above, you can already reach on your own. This one you cannot get anywhere else.
Underneath the cited pages sits a population substrate, built on two clearly separate figures. The atlas map spans 18,477 research dog genomes across hundreds of breeds, drawn from three open research releases (CanVAS, Darwin's Ark, and NHGRI) and embedded by genetic similarity. Per-dog genotype analysis runs on the 14,478-genome CanVAS release. Alongside both sit breed-level carrier frequencies that carry their sample size and confidence on every value. It is a real map of the real population, not a model's guess about it.
It is honest about its own limits. It supports genomic relatedness, breed structure, and diversity position. It does not predict disease from the embedding, because we tested that and it collapsed to chance once relatedness was controlled, so we do not ship it. For a comparative-medicine or population-genetics question, this is the part that gives an answer a denominator: not just "this variant exists," but roughly how often, in which breeds, with what confidence, and cited.
And where a cohort has been followed for life, the map gains a time axis. The Golden Retriever Lifetime Study has tracked 3,044 goldens from puppyhood, and its published cohort numbers give real incidence and age-of-onset for the cancers and conditions of the breed, cohort-framed and cited.
The dog is a model for human disease
Because dogs and humans share most of their genes, a canine disease can often be read against the human disease it models. Sniff makes that link explicit and cited, in both directions, and it stays honest about the cases where the translation does not hold.
Canine Degenerative Myelopathy is associated with a variant in SOD1, the same gene implicated in a form of human ALS. In dogs the variant acts as a risk factor with incomplete penetrance, not a guaranteed sentence, which makes DM a naturally occurring model of SOD1-related motor neuron disease. Sniff shows the dog disease, its human analog, and the shared biology, each linked to its source, and framed as a model of the human disease, never asserted to be it.
See it: Degenerative Myelopathy and human ALS →
The same logic runs through cancer. Naturally occurring canine osteosarcoma is a conserved genetic model of human, especially pediatric, osteosarcoma, driven by the same core tumor-suppressor losses. Sniff shows the shared somatically altered driver genes, from peer-reviewed cohorts and cited to each. These are tumor alterations reported as cohort frequencies, kept separate from germline carrier status, and never a prediction about any individual dog.
Why you can trust it
Every rule below is enforced in the build, not promised in a mission statement.
- Cite or abstain. If a claim cannot be sourced, Sniff does not make it.
- Sample size and confidence on every value. Below a floor, a number disappears and says "insufficient data, n=X."
- A variant of unknown significance is shown as uncertain, never dressed up as a risk.
- The human analog is a model of, never an identity. The dog's disease stays the subject; the human disease is the analog it models.
- A confident wrong answer is the single failure we work hardest to prevent.
Who it is for, and an invitation
Comparative-medicine and One Health researchers, geneticists, veterinarians, evidence-based breeders, and the standards bodies defining the field's canon. One substrate, many front doors.
Sniff is free and open, and it will stay that way. Use it, cite it, and above all, tell us where it abstains when it should not, or overclaims when it should not. That is the bug we care about most, and the fastest way to make it better for your work.
The end state is open, cited public infrastructure for animal genetics and its bridge to human medicine: the kind of resource a researcher, a vet, a breeder, and an owner can all lean on, and none of them has to pay for. Instantiated on the dog, architected for the tree of life. Built on the field's canon, kept honest by construction, and given away because that is the way it should be.
Built and maintained by Matt Gehring, an independent researcher (ORCID 0009-0001-9531-2861), operating as Candor Systems LLC. Corrections and source suggestions are genuinely welcome at [email protected].