23andMe and AncestryDNA Ethnicity Tools
The recent admixture analysis tool at 23andMe “Ancestry Composition” and it’s counterpart at AncestryDNA the “Genetic Ethnicity Summary” has generated a lot of discussion about why none of these types of ethnicity calculation analysis seem to turn out close to correct. Here’s one reminder nipped from my Irish cousin, Gerard Corcoran’s Pinterest blog about how fluid those European country borders are. Not only can societal demarcations alter the free movement of genetic material, it can alter our perception of ‘where’ a segment that is German, for instance, is from. Do you think of pre-WWI Germany or Germany as it was 1000 years ago or Germany as it stands today when you imagine your ancestors lives?
But what if 23andMe’s customers thought of a different time frame when they listed the location of their four grandparents. What if they listed their birthplaces according to the borders at the time they emigrated to the US or Canada? What does Germany actually means. If they put German for all four grandparents birth locations, they may have been used in the German population that 23andMe built from their customer database but what if their grandparents are really from Prussia?
23andMe Ethic Accuracy
In the following chart, Recall means the number of segments that originated in — for instance — Italy that 23andMe’s algorithm can properly categorize. Think of it as how large the pile of uncategorized segments is after as many segments as possible have been categorized. So if you have 10 British/Ireland segments and 23andMe only finds 3 of them, you have 70% left behind (or 0.7). The Precision is what percent of the segments that 23andMe labels is labeled correctly. So if they show you with 10% British/Irish then1% of those segments are improperly labeled. Note that none of the Precisions are actually perfectly 1, but my excel sheet insisted on rounding them. For more on Recall and Precision, please see my previous post.
|0.98||East Asian & Native American||1|
|0.32||British & Irish||0.9|
|0.07||French & German||0.74|
|0.87||Middle Eastern & North African||0.92|
My Ethnicity Calculation
To graphically demonstrate the problems associated with historical ethnicity determination, here is a peek at my ethnic breakdown at all three companies. As you can see, FTDNA gives me the big shrug. AncestryDNA seems to go too far back in time and is finding Nordic invaders/Seamen and 23andMe is dancing in between. I’m not sure if they got the Scandinavian right and worry that they have deliberately lowered that percent (they admit they did) so it fits customer perceptions.:
|Genetic Ethnicity Summary||Ancestry Composition||Population Finder|
|See British Isles||Western Europe||See British Isles||93.19%|
|2%||Error/Unknown||0% or 0.7% Nonspecific European||2.15%|
I think the key to improving 23andMe’s tool is localizing the populations more and basing them on the map 500 years ago. Which brings us back to this fun video from TheMuzikkeyfi on YouTube. The music that this dynamic map of Europe evolves to is “Dream is Collapsing” by Hans Zimmer for the movie Inception.
Evolving Europe Video
Animation Credit: TheMuzikkeyfi on YouTube
Music Credit: “Dream is Collapsing” by Hans Zimmer for the movie Inception.
Credits: Clan Heber aka Gerald Corcoran on Pinterest