CanVAS.

Why CanVAS exists

Canine genomics has a sharing problem. Not the kind people think. The data is shared. The papers are published. The supplementary files are on bioRxiv. The problem is that none of it is comparable.

Fifteen different research groups, fifteen different genotyping arrays, fifteen different reference genomes, fifteen different probe naming conventions. A study from 2016 used CanFam3.1. A study from 2022 used CanFam4. One paper's "BICF2P1234567" is another paper's "BICF2P1234567_dup1" because the strand orientation flipped between chip versions. The Golden Retriever Lifetime Study cohort cannot be compared to the Hayward 2016 cohort without weeks of bioinformatics surgery, even though both are publicly available, even though both are answering related questions about the same species.

The result is that canine genetics, as a field, has been studying dogs in fifteen rooms with the doors locked. Every group analyzes its own cohort. Cross-cohort meta-analyses are rare because the lift is enormous. Rare-variant discovery, which requires sample sizes that no single cohort delivers, is largely impossible.

CanVAS is the work of unlocking those doors.

What harmonization actually means

Imputation is the multiplier. It is the move that turns 77,215 directly measured SNPs into 9.7 million inferred variants of comparable quality. It is also the move that lets a sparse, decade-old genotyping chip stand alongside a modern whole-genome study in the same analysis.

What 15,451 dogs at this density unlocks

Sample size and variant density compound. CanVAS gives the field both at the same time, for the first time.

The dog space CanVAS covers is wider than any prior single resource. 375 breeds. Village dog populations from multiple continents. Dingoes. Wolves. Coyotes. The aggregate is a population-scale view of the canid genome that resolves both within-breed substructure and the relationships between breeds, breed groups, and the wild canids that share their ancestry.

Rare variants change the questions the field can ask. Most genome-wide association studies in dogs have been powered for common variants because that is what the chips capture. The 3 million rare variants CanVAS recovers through imputation move parts of canine GWAS into the same statistical regime as modern human genomics, where rare variants with strong effects are the frontier of disease genetics.

The Brundage team validated CanVAS by recovering known population structure and genome-wide runs of homozygosity, both of which behave as expected if the harmonization and imputation are correct. The atlas reproduces breed-level inbreeding patterns that match the existing literature. The infrastructure works.

What CanVAS contributes to the Sniff Atlas

Every dog in the Sniff Atlas is a CanVAS dog. Without CanVAS there is no Sniff Atlas.

The 14,478 dogs visible in the Sniff Atlas today are the subset of CanVAS that passes our additional quality control. From those 14,478 dogs across 342 breeds, the entire layout of the atlas is computed. The PCA-256 embedding that defines genetic similarity. The 3D UMAP projection that gives every dog a spatial address. The per-breed centroids that anchor the breed-isolation view. The neighborhood placements that tell a newly uploaded dog which research dogs they sit closest to. All of it derives from CanVAS genotypes.

When the atlas adds a new study or cohort in the future, it will either be by ingesting an updated CanVAS release or by harmonizing a new source to the same reference. The Sniff Atlas is downstream of CanVAS. That is by design. The science has to live on a single shared substrate or the atlas stops being trustworthy.