Why Admixture Calculators Have To Be Interpreted Cautiously

[Angela]

This was put together collaboratively by Graham Coop, Iain Mathieson, Joe Pickrell, and I don't know how many others:

How Not To Use Admixture:

Obviously, people have done all of these things.

It's something to keep in mind when looking at Admixture results in papers. Make sure you read the methodology section carefully. When it's done by amateurs who don't publish the methodology it's even more important to be cautious in accepting the conclusions. You should be most cautious of all when it's all based on modern populations.

Of course, formal stats are not immune from problems with methodology. I'd like to see something similar written for them.

 

[Auld Reekie]

I just tried phasing for the first time with Gedmatch using my mothers and my results trying to see what my late fathers DNA might be. He had some 2-4 sub-Saharan African and 6-15 South Asian and the rest was northern and central European. There isn't any African at all in my sister or me though there is some South Asian. I'm not saying the sub-Saharan African couldn't be in my father but could that be archaic DNA or noise? I'm trying to figure out where it comes from or what it all means. I don't want to go running to tell my family something that might not be true. I also ran my mother's phased version against the one I uploaded and they were both different. The phased one was more northern Italian and French and the other was more central/southern Italian and Sicilian. All the weird bits of her Amerindian, Pygmy, Armenian and Siberian were gone also.

 

[davef]

Let's elaborate....
Suppose someone who is of 100 percent german ancestry gets "Dutch" as his top population match and German as his second closest. I chose Dutch because this population is very close genetically to the Germans.

Now we should assume there's way more Dutch samples than german samples. If we had more german samples, we would likely run into a German who's genetically more similar to the testee as the closest Dutch in the original sample pool so if we run the test again, German would show up on top, Dutch second. Is this one way the sample pool can skew things? I'm just making guesses here.

Further, do these tests look for IBS matching? Obviously it can't be IBD. I know IBS means short segment sharing (however you want to define short). So if someone from France who's family has always been in France for millenia gets 2 percent Syrian, that just means that 2 percent of that persons genome consists of genes that are most often found in the average Syrian in the sample pool, correct? The connection could be from the Stone Age, Roman visitations, etc. The match is purely IBS so say we have a really old population ancestral to both the French and Syrians and we have this sequence of genes: ABCD and the letter position is the gene position. People migrate and over time someone from France gets the sequence ASED and someone from Syria gets AYRD. Now for these two samples to be IBD, there needs to be a sequence at least 3 genes in length that matches between the two but in this case no such sequence is found but we do have them matching in the first and fourth position in that these positions have genes A and D in common, an IBS match. ADMIXTURE will account for this commonality. Is my understanding correct?