gnomAD.

To know whether losing a gene matters, look at how often people are missing it, and get away with it.

gnomAD, the Genome Aggregation Database from the Broad Institute, pools the genomes and exomes of more than 800,000 people and asks a simple question of every gene: how much natural variation does it tolerate? For a gene that the body cannot do without, loss-of-function variants are quietly selected away, so almost none show up in the population. For a gene that tolerates being switched off, they turn up freely.

gnomAD turns that into a number. Its constraint metric, LOEUF, measures how strongly a human gene is selected against losing function. A low value means the human population rarely tolerates that gene being broken; a high value means it does. It is one of the most widely used signals in human medical genetics for judging whether a gene is likely to matter.

It is the human context waiting on the far side of the ortholog.

When a dog gene resolves to a single human counterpart through Ensembl Compara orthology, Sniff reads gnomAD's constraint for that human gene and states it plainly on the gene page: in people, this gene rarely tolerates loss of function, or it appears to tolerate it. That is one honest step of comparative context, from a dog's genetics toward what human medicine has measured.

We hold two hard lines. It is gene-level and about the human counterpart, never a verdict on the dog's own variant. And constraint measures intolerance to loss of function only: a gene that tolerates being switched off is not therefore unimportant, since many such genes cause disease through other mechanisms. So the surface stays symmetric, a tolerant gene gets a stated, cited line just as a constrained one does, and where the human ortholog is not one-to-one, the whole layer abstains.