Some clues about the sensitivity of 23andMe's ancestry analysis

23andMe labs recently added a "Native American Ancestry Finder":
Do you have Native American Ancestry? This feature scans a person's Ancestry Painting for distinctive signatures that indicate a Native American ancestor up to five generations in the past. It also takes into account the maternal and, if available, paternal lines, looking for Native American ancestry at any depth along those two branches of the family tree.
The "five generations" number is the minimum number of generations, based on 23andMe's simulations, after which any trace of American Indian ancestry may fall below the threshold of detection.
How can we be sure this person has no genetically Native American ancestors in the past five generations, and what Is the likelihood prior to that?

To address this question we start with the observation that people who identify themselves as Native American exhibit fairly consistent Ancestry Painting proportions of about 75% Asian and 25% European, plus or minus 10%. The reason: Native Americans descend from a small number of people who crossed the Bering land bridge from Siberia more than 14,000 years ago. In Ancestry Painting, Siberians and people from many other central and northern Asian locales tend to have a roughly three-to-one proportion of Asian to European DNA simply because they lie geographically and thus genetically intermediate between the Asian reference population, which consists of Japanese and Han Chinese individuals, and the European reference population, which consists of Americans of northern European descent.

Next we consider what would happen to that three-to-one Asian/European proportion over the generations if a Native American and a partner of all-European descent had a child who then reproduced with another all-European partner, and so on. In such a case, the amount of Asian DNA in each successive generation's Ancestry Painting would necessarily diminish, until at some point it disappeared altogether.

To get an empirical idea of how this process works, we created simulated Ancestry Paintings for 1,000 people with one Native American and one European parent, then 1,000 people with one Native American parent and three European grandparents, and so on down the line. Then we ran all these simulated individuals through Ancestry Painting, and looked at the range of Asian and European DNA percentages for each kind of relationship. (We also looked at the African percentages for these simulations, but those were nearly always zero, or a trace at most.)

We found that it takes at least five generations after the appearance of a Native American for the percentage of Asian (orange) DNA to reach zero. So we're able to say with confidence that a person with an all-European Ancestry Painting (actually, 99.74% European or greater) did not have any genetically Native American ancestors in the past five generations.
Since the promised white paper on Ancestry Painting has yet to appear, this seems to be the best information available on the accuracy and precision of 23andMe's admixture analysis. Presumably, Asian and particularly African ancestors would be detectable even farther back.


notuswind said...


What are your preliminary thoughts on 23andMe's ancestral painting analysis?

As a customer of 23andMe's flagship product I'm particularly curious about this.

n/a said...

Short answer: I like it. Especially in combination with Global Similarity Advanced View.

I would like to see more technical details, but it's clear that of the commercial autosomal ancestry testing options available, only 23andMe consistently gives plausible results.

Decodeme test more markers, but only use around 18000 SNPs in their admixture analysis. 23andMe supposedly use all 500000+ SNPs for Ancestry Painting. Probably more importantly, Decodeme's algorithm systematically overreports non-European ancestry in Europeans, while the overwhelming majority of whites are reported 100% European by 23andMe. I also expect 23andMe will devote more resources to improving their offerings in this arena than Decodeme.

(The other players in this market don't even deserve mention: DNAprint was crap and is now thankfully dead. DNAtribes is even more useless DNAprint.)

Anonymous said...

history is only biography.

notuswind said...


Thank you for the response.

My feelings about the analysis are also similarly positive. What I found to be most impressive was that 23andMe's Global Similarity Advanced View was able to correctly identify what I understood to be my dominant ethnic background (based on family genealogical records). All that from a spit sample!

Of course, I'll be interested to read your summary of the technical details once they become available to you.