How do DNA ethnicity results for identical twins compare at AncestryDNA, 23andMe and MyHeritage? Read the answers here, along with a bonus question about 23andMe traits.
Somebody finally did it. A research group has tracked the ethnicity results of 21 pairs of identical twins at three different major companies (AncestryDNA, 23andMe, and MyHeritage DNA) to get a better picture of the consistency of results. Earlier this year, they published the results in the American Journal of Medicine (Epub 2019 Jun 15).
Up until this point, there had only been scattered reports of single sets of twins (see one here), so this is a welcome piece of information to get a clearer picture of what’s going on with these ethnicity reports. What do you think they found? Did each twin in the pair have exactly the same ethnicity proportions, or something wildly different for each? Either way, it’s great that there is a study on record with some hard numbers on how comparable these tests are for the same DNA.
Ready to see what they found?
First, let’s set a ground rule: identical twins have essentially identical DNA, for the purposes of consumer DNA testing. A study in the Journal of Medical Genetics (2014 Jan; 51(1): 28-34) estimated that pairs of identical twins have on average 420 single-point differences in their genomes due to mutations that occur very early in fetal development. In a genome of 3.5 billion bases, 420 base differences is a very small amount, and further it’s unlikely that the bases where they differ are the population-informative SNPs the companies use for ethnicity testing. All that to say: using identical twins for this test of concordance of ethnicity estimation is a solid approach.
Identical Twins and DNA Testing for Ethnicity
In this study, the research team asked two different questions. First, they measured how similar ethnicity estimates were when the 21 pairs of identical twins were tested at the same company. Secondly, they disregarded twin status and just looked at how similar ethnicity estimates were for the same person at different companies for all 42 study participants. The results are telling:
First, for pairs of twins analyzed at the same company, consistency for ethnicity estimates is high. This is great news. Internal consistency for a company gives consumers confidence that the lab processes are reliably executed and that the computational models draw out essentially the same genetic trends from the same DNA when challenged in independent tests. That’s good. 23andMe demonstrates somewhat of a dip in internal consistency, with an average measure of 94.5%. This may indicate some looseness in the algorithm that predicts ethnicity. This is not entirely unexpected due to the nature of these algorithms which by design handles the highly complex genetic data in a way that does not produce 100% agreement for the same person. All considered, I’d say this is a strong showing for all three companies.
Numbers in the second part of the study may leave some of us stroking our chins. Hmmmm. They’re not as high as we’d like to see, are they. When the ethnicity estimates of 42 people were compared between the three companies there is a marked drop in consistency. The best concordance is seen in the tests for the same person between AncestryDNA and 23andMe at 84.1%.
The bedrock of ethnicity tests are the reference panels that are used to compare the test-takers’ DNA and make predictions about their ethnicity proportions. Each company uses its own proprietary collection of DNA samples to make up their reference panel. AncestryDNA and 23andMe demonstrate the best consistency, and this could be in part because their reference panels are the largest. It’s possible that their larger sampling enables them to detect genetic trends more consistently than when a reference panel is based on smaller numbers and less thorough sampling of the genetic character of a region. The reference panel used by MyHeritage is ½ the size of that used by 23andMe, and ⅓ the size of AncestryDNA’s panel. This could be one factor among many that is producing results with lower concordance.
All in all though, to me, 84% agreement between AncestryDNA and 23andMe can actually be seen as pretty darn good. Think about it: these companies don’t use the same SNPs; they have a completely different approach to their computational models; and their reference panels are totally unique. This measure of concordance shows that each company is independently detecting genetic trends that—although not perfectly—in large part DO agree with one another. They are independent witnesses producing similar findings, using lenses that are considerably different from one another. As a consumer, that’s something that I want to pay attention to.
The greatest disparities in ethnicity estimates are seen for comparisons with MyHeritage: 23andMe was on average 69% similar, and AncestryDNA was only 52.7% consistent. None of the comparison numbers between companies are perfect, but this study suggests that MyHeritage is consistently producing results that are outliers from the two frontrunners of the industry. Many of the factors that can lead to differing results between companies are examined in detail in my blog series exploring ethnicity estimation, if you’d like to brush up on how the companies generate their results in the first place.
Bonus Question: 23andMe Traits
Another aspect of this study explored the participants’ agreement with traits predicted by 23andMe based on their genetic test.
The levels of agreement seen between the predicted traits and the clients’ actual experience are all over the map. Clients agreed 90.2% of the time with the genetic prediction for having a cleft chin, and concordance plummets as low as 34.1% for the deep sleep trait, with all other traits falling in the spread between. The traits that don’t seem to predict well may demonstrate that the genetic factors that produce these characteristics are not yet fully understood, while some of the higher performing traits show that predictions can be pretty meaningful. As a consumer, understanding how strongly the genetic prediction is associated with actually developing the trait gives us a better sense of how much weight we can lend to the results on a trait report.
This is just one small study of the compatibility of results from direct-to-consumer ancestry testing. The sample size of 21 sets of twins is modest, the study participants were primarily white non-Hispanic women, and the authors themselves point out some of the limits to interpreting their results, including that comparisons involving 23andMe may be inflated for consistency. But it still contributes to our understanding of how all of these puzzle pieces fit together that we get from different companies (and perhaps whether any of the puzzle pieces should be thrown out altogether).
The authors’ conclusion is that consumers should not be misled about the usefulness of these services. Do you agree? With the data from this identical twin study in hand, how you interpret your own ethnicity estimations is still up to you.
Did you know? Our Autosomal DNA quick reference guide isn’t just for beginners: it’s got a detailed table comparing the ethnicities reported by the big DNA testing companies. It may help you find the test that reports a certain ethnic origin in the most detail!
And for deeper exploration into what your ethnicity resultsmean, take a look at our free guide on DNA ethnicity estimates.
I want more info about that free guide!
