Thursday, 6 April 2017

Why Are Your DNA Ethnicity Results Unexpected: Analysis

Just before Christmas I received an email from Living DNA to tell me my results would be delayed. The first sample had returned too many no call results.
When the results did arrive it was in advance of the original date promised. Good job Living DNA.

It made me wonder, how many no calls is too many? Do all the SNPs used in the analysis have to be called, or is there some flexibility? How many is enough? How much will the analysis be impacted if some are missing?

Some insight comes from the two sets of results, analysis of the same person by 23andMe at different times.

European
98.2%98.3%
British & Irish55.7%56.7%
French & German3.1%3.5%
Scandinavian0.1%0.1%
Broadly Northwestern European13.0%12.0%
Ashkenazi Jewish18.8%18.8%
Iberian0.0%0.9%
Italian0.4%0.4%
Broadly Southern European4.5%3.7%
Broadly European2.3%2.0%

The results are close, all within 1 per cent. The companies should stop giving a false sense of how accurately the percentages reflect ancestry, at least by rounding out the decimals?

Another analysis factor is the reference databases used by the various companies. While there are some databases openly available the companies add data from other research including their clients results. The recent changes in Family Tree DNA's myOrigins assignments, which were introduced with minimal explanation - "smaller trace-percentage results" - "newly refined European, Middle Eastern and Native American clusters," illustrate the impact.

If you tested with more than one company adding to the confusion is the the different ways the companies segment and characterise regions. Reminiscent of the saying that a person with a watch knows the time, someone with two is never sure.

This is the final post in the Why Are Your DNA Ethnicity Results Unexpected series. If you missed the others, or want to read them again, they are:

Misattributed Parentage
Brick Walls
Recombination


No comments: