Geno2 users showing unexpected Denisovan ancestry

I have been excited to hear in the last few days from several readers who have gotten results from the new Genographic Geno2 genotyping chip. One aspect of the result reporting is a person's estimated proportion of Neandertal ancestry, which is a simple percentage. This is like the report from 23andMe, and should be a pretty straightforward estimate given a model of Neandertal-human genetic similarity from complete genomes.

Another aspect of the Genographic results is an estimated proportion of Denisovan ancestry. This might seem a bit surprising, as for most participants in the project who lack Polynesian or Melanesian ancestry this proportion should be extremely low. I've written about Denisovan DNA similarity with living peoples a few times ("Denisovan DNA in the islands, and an Australian genome", "How widespread is Denisovan ancestry today?"). Based on the science published to date, I would have expected the Geno2 calculations just to confirm the very low ancestry estimation found in last year's research based on genotyping Asian and Australasian populations.

So I have been extraordinarily surprised to see that people are getting Geno2 results with up to 6% Denisovan ancestry!

What gives? None of my correspondents so far has anything other than European self-reported ancestry, making it seem very unlikely that have substantial Denisovan ancestry.

The first time I heard from a reader with this result, my immediate reaction was that there must be some problem with the algorithm. This one in particular wouldn't be to hard to get wrong considering the rarity of whole genome evidence from populations known to have substantial Denisovan ancestry. Or possibly, some problem with an individual's genotype chip data might trigger the algorithm to look more Denisovan. For many loci that vary among humans, Denisovans are very unlikely to have the derived human variant; so an individual with an unusual proportion of ancestral homozygote loci might look Denisovan in a human-Denisovan comparison.

However, this is all speculation without knowing the details of the Genographic analysis. And as I hear from more people with varied results, I am having trouble thinking of how data errors could be patterned. It's a tough one to think about because of the unique aspects of the Geno2 chip, and until I've gotten a feel for results from that platform compared to other datasets I probably won't have a solid idea.

I should point out that if there is a problem with the algorithm underlying ancestry prediction from Denisova, it almost certainly affects the Neandertal ancestry estimate also. The estimation from both these ancient genomes involves the same procedure, although with Denisovan DNA it requires subtracting out the DNA similarity with Neandertals first.

So I would be interested in hearing from anyone who is surprised to find that they have Denisovan ancestry. My preliminary assumption it that the result is spurious but I'll try to figure out if there is a possibility of some Denisovan fraction beyond what has been shown in published work.