A look at the whole breed, one dot per dog, and a question: how few dogs could we DNA-test and still learn about most of them?
—
of dogs are as related as half-siblings (counting all shared ancestors) to a dog we tested
Tap or hover a dot to learn about it
More options
dogs shown
distinct families
average inbreeding
with inbreeding above 12.5% (a half-sibling mating)
Table view
Family
Dogs
Average inbreeding
Median birth year
How this was made
What "covered" means. A dog is covered when a sequenced dog is at half-sibling level or closer (relatedness 0.25 or more), so its genome can be filled in from that relative. The window of birth years 2012 to 2020 stands in for the dogs alive and breeding today. "Who counts" narrows the dogs measured to those born in a recent stretch of it; the sequenced dogs can come from anywhere in the window for the one-at-a-time strategy, and from the counted dogs for the groups. The most recent years are thin because registrations lag.
The reference genome. The Great Dane reference sequence came from a female Great Dane named Zoey (Halo and others, PNAS, 2021, “Long-read assembly of a Great Dane genome…”). The paper does not give her pedigree or say why she was chosen, so she cannot be placed on this map.
The pedigree. A pedigree database export of show-bred Great Danes born 2012 to 2020, plus the ancestors of those dogs, with a few close relatives of ours added from our own records. Dogs with both parents recorded are kept. Our own litters, born or planned after 2020, are added as one node per litter.
Relatedness. For every pair of dogs, the expected share of genes identical by descent, computed from the whole recorded pedigree (up to 77 generations back). The same calculation gives each dog's inbreeding. Because it counts every recorded generation, these figures run higher than the 10-generation figures on the pedigree explorers.
Families. Dogs are grouped by average-linkage clustering on relatedness. Groups above 600 dogs are split again. Each group is named for its dominant coat.
The map. Families are placed as islands, pushed apart or pulled together by how many close relatives they share. Inside an island, each dog is a particle held by springs to its four closest relatives. The layout is settled ahead of time, then the physics keeps running in your browser, so you can drag and shake it.
Splitting into k groups. Like k-means, but the center of each group must be a real dog and distance is kinship. Start with one group. Each step divides the group where two centers raise average relatedness the most, with each family moving as a whole so groups line up with the islands. A group's center is the dog at half-sibling level or closer to the most of its members, and that dog is the one to sequence.
Sequencing. A dog counts as covered when a sequenced dog is at half-sibling level (relatedness 0.25) or closer. "Best next dog" greedily picks the dog that covers the most uncovered dogs. "One per family" takes a stand-in from each family, biggest first.
Keeping it small and anonymous. The page carries only a position, birth year, coat family, family number, and inbreeding figure per dog, in a shuffled order, with the links packed as 16-bit numbers. No names, registrations, or kennels are included, except for our own eight nodes.
Want the full analysis? The Kinship Study has the methods, 140 years of breeding choices, and data quality. The sequencing proposal says what we would do with it.