Admixture, heritage, ethnicity estimates

[1geewong]

It seems very common for people to show a high level of interest in their admixture/heritage/ethnicity results from their DNA testing. "My pie chart says...!"; you'll recognize that from the AncestryDNA advertisements. A challenge in Chinese genealogy is that the models used to estimate ethnicity by the major testing companies focus on European ancestry. Some of the minor testing and analysis organizations may focus on Asian ancestry but offer no explanation for the basis of their analysis.

I've found a really nice explanation of the admixture estimation models available at GEDMatch (http://genealogical-musings.blogspot.com/2017/04/finally-gedmatch-admixture-guide.html ).

This blog not only explains the origins and weaknesses of the general process of admixture estimation models but also provides good guidance on which models are best to use for a variety of suspected ethnicities. If you aren't up for reading the blog and its pertinent links, I think the bottom line for Asians is to use admixture model "Dodecad V3"; Asian mixed with other backgrounds use "Dodecad K12b". At the end analysis, check the Oracle and Oracle 4 buttons for ethnicity estimates (lower the number, the better the ethnicity match).

A good demonstration of the imprecision of admixture models can be found by running the analysis on GEDMatch called "Chromosome painting". It graphically shows the various ethnicities that share the same DNA-types found at each site on the chromosome tested. It's a very imprecise analysis.

[jeremychin]

1geewong,

Is this what you are referring to?


If so, it is the complete opposite of my 23 and me result of 96% Chinese. Wegene is also more in line with 23 but with a slightly higher SEA %.

[1geewong]

Hi Jeremychin,

Those are the analyses I'm referring to. I'm also commenting on the seemingly wide range of results; their imprecision. The admixture analysis is an evaluation of a probabilities based on models .... the models are deep science, above my "pay grade", and difficult or even impossible to correlate with result we have access to (i.e. what model is "X" organization using? what is its validity? what population data was used to develop the model?)



I've tried to post an image of chromosome painting from GEDMatch; hopefully it will show up. If so, the image is a small segment of one chromosome. The various colors indicate different "ethnicities" that all share that same DNA type found at that point on that chromosome. In many places on this person's chromosome, the DNA type is only found in one ethnicity. In other places, several ethnicities share that DNA type to varying degrees - some decision has to be made of which one contributed to that person's DNA. That decision(s) will impact the output result you are given but may or may not accurately reflect the person's actual heritage.

[ceceproud]

Dec 11, 2017 9:46:37 GMT -5 1geewong said:

Hi Jeremychin,

Those are the analyses I'm referring to. I'm also commenting on the seemingly wide range of results; their imprecision. The admixture analysis is an evaluation of a probabilities based on models .... the models are deep science, above my "pay grade", and difficult or even impossible to correlate with result we have access to (i.e. what model is "X" organization using? what is its validity? what population data was used to develop the model?)



I've tried to post an image of chromosome painting from GEDMatch; hopefully it will show up. If so, the image is a small segment of one chromosome. The various colors indicate different "ethnicities" that all share that same DNA type found at that point on that chromosome. In many places on this person's chromosome, the DNA type is only found in one ethnicity. In other places, several ethnicities share that DNA type to varying degrees - some decision has to be made of which one contributed to that person's DNA. That decision(s) will impact the output result you are given but may or may not accurately reflect the person's actual heritage.