Monday, August 18, 2014

My Autosomal DNA Quandary Persists

I wrote My Autosomal DNA Quandary a year ago, and my quandary still persists.

In a nutshell, my quandary is that the three autosomal test results I've taken do not match each other, and only one matches my perception of my genealogical ethnicity.

According to my genealogical research, based on the estimated origins of my 256 6th great-grandparents (I had to guess at some of them), is in these regions:

68%  British Isles
24%  Germany
1%  Holland
1%  France
6%  Unknown (perhaps some are French, or English, or Native American)

Compare that estimate with the results provided by the three autosomal DNA tests taken to date:

1)  The AncestryDNA test results:

The AncestryDNA ethnicity estimate is:

*  66% Europe West
*  18% Ireland
*  9% Great Britain
*  3% Scandinavian
*  1% Italy/Greece
* <1% Iberian Peninsula
*  <1% Pacific Islander
*  < !% Native-American

2)  The FamilyTreeDNA test results:

The FamilyTreeDNA ethnicity estimate is:

*  45% Scandinavia
*  32% Western and Central Europe
*  19% Southern Europe
*  4% British Isles

3)  The 23andMe test results are (Speculative estimate):

The 23andMe ethnicity estimate is:

*  46.7% British & Irish
*  24.7% French & German
*  1.9% Scandinavian
*  23.5% Broadly Northern European
*  1.2% Broadly Southern European
*  1.2% Broadly European
*  0.5% East Asian & Native American
*  0.1% Middle Eastern & North African
*  0.1% Unassigned

4)  As you can see, the ethnicity populations are not exactly the same between the three test companies.  

*  If you combine British and Irish, then 23andMe says I'm 46.7%, AncestryDNA says 27%, and FamilyTreeDNA says 4%.  [Compared to 68% from genealogy research.]

*  FamilyTreeDNA says I'm 45% Scandinavian, AncestryDNA says 3%, and 23andMe says 1.9%.   [Compared to perhaps 1-2% from genealogy research.]

*  If you consider France and Germany to be representative of Western and Central Europe, then the AncestryDNA says 66%, FamilyTreeDNA says 32% and 23andMe says 24.7%.   [Compared to perhaps 28 to 32% from genealogy research.]

*  Only 23andMe provides non-specific ethnicity results, saying I am 23.5% "Broadly Northern European" (which could be Scandinavia, Western and Central Europe).  AncestryDNA and FamilyTreeDNA don't provide any "broad" results.

So do you see why I am still confused?  Also, AncestryDNA and FamilyTreeDNA have changed their ethnicity estimates in the past year or two.  

5)  My understanding for the disparity between these results rests on three factors:

*  The estimates are from DNA test results in reference populations in the specific areas.  Very few of these European populations are homogenous - we know that the British Isles have had significant immigration over the centuries from Scandinavia (Viking, Norman), Iberia, and western Europe (Angle, Saxon, etc.).  Perhaps that explains the British Isles and Ireland ethnicity disparities.

*  The autosomal DNA in our chromosomes comes from ancestors from thousands of years ago, not just 7 generations ago.  

*  Each testing company uses the same chip, but tests different numbers of SNPs, or gives more importance to certain SNPs.

Obviously, I am not an expert in this - just a confused customer and user.

If the tests all agreed, and were really different from my genealogical estimates, then I would think that I have a serious genealogy research problem.  But 23andMe seems to match my genealogical estimates pretty well, especially if I apportion the "broadly Western European" segment to the British Isles and France/Germany.

One concern is that each of the DNA testing companies seem to claim that the results provided are definitive for each person.  Obviously they are not, at least in my case.  We saw a lot of DNA test results on the Finding Your Roots TV series on PBS last year, and I anticipate we'll see more this year when the 2014 season starts in September.  I think they are using 23andMe for those tests, but I'm not positive.  

There was quite a bit of helpful discussion in the comments to my earlier post My Autosomal DNA Quandary  (posted 22 August 2013).  Since then, I have worked a bit in GEDMatch to try to find chromosome matches.  I have not investigated DNA Tribes to date.

Some questions:

*  How well do your test results match across companies?

*  What are the uncertainty bounds for each test company?

*  What am I missing?

*  Should I just accept that the reference populations and methodology are different between test companies?

*  Will the estimates from the testing companies be improved again and again over the coming years?

Reader comments will be appreciated!

The URL for this post is:

Copyright (c) 2014, Randall J. Seaver


Judy G. Russell said...

The only thing you're "missing" is that these estimates are only truly accurate at the continental level, not at the country level. In other words, the specific percentages below Europe, Africa, etc., are cocktail party conversation and not scientific "reality."

Unknown said...

Hi Randy,
Thanks for the post. I know in my logical mind that the ethnicity estimates are the least reliable piece of data in the DNA portfolio, but I have great hope that they will improve in their accuracy as samples increase. Having said that, I have tested with Ancestry and FTDNA and my results are remarkably consistent...

Debbie Kennett said...

The point you need to remember about these ethnicity labels is that they represent genetic clusters identified by computer algorithms from small datasets with limited numbers of reference populations. The companies decide how many clusters they want to use and then assign arbitrary geographical labels to these clusters. See the autosomal section of this article which explains the limitations:

This blog post provides a good explanation of how the STRUCTURE program which is used by all three testing companies actually works:

Ginger Smith said...

I agree with everyone who has posted comments so far. It is my understanding that each company has their own reference populations. AncestryDNA is using Sorenson's database in addition to new populations being studied, for example. Each company has a different number of samples from each geographic area. I don't have my notes in front of me, but I remember, for example, there is one company that has a much better reference population of Africa. So, if you are African, you would get more valuable results by testing with that company. I will look that up when I get home tonight.

David said...

Randy, like you, my ancestry is almost all European, though with the vast majority being German. I have also found 23andMe's results to be most in line with my paper trail and it will be the test of choice when I have other family members tested in the near future.

Ancestry's test was OK, though with a little bit of head scratching involved. I think there is still room for improvement.

I had issues with FTDNA on some transferred results and will not use them again.

Kitty Cooper said...

Randy now that GEDmatch is back up try the various Admix calculatos there. Start with Eurogenes K13 and then try the rest. I did a presentation on GEDmatch and included many slides on the ancestry compostion calculators. You can get the link to the slides from my blog post on this.