Thursday, 28 April 2011

Searching for Native North Africans, Part II: EastAfrican-Americans

I'm short on time, so I'm only presenting for now the Masai restricted-pole results and some experiments with 53 African-American samples.

I used either the full set or just five individuals as poles in each run. I didn't use 24 vs 5 since I assure you for the following analysis the behaviour of poles comprising all individuals vs 24 is basically the same.
Fst: 164
This is the straightforward result against which all other runs should be compared. Yoruba are historically a good proxy for the West-African parent population of African-Americans. White-Utahns are obviously a near perfect proxy for the other ancestor population. There is also a little bit of Amerindian but I decided to ignore it since it's so small, and after realising in the Mexican run that small component poles can act quite strangely in such experimental set-ups.

Here's what happens when the Yoruba pole is reduced to just five individuals:
Not much change. Yoruba here are representative of West African ancestors and a valid full-set pole.

Now all Masai as the pole.
Fst: 108
Such discrepancies suggest what's already well known. The Masai are not good proxies for the parent population of African-Americans.

Using five Masai as the African pole:
Fst: 123

Genetic distance between components is now considerably closer to the actual one, and much of the divergence is due to unavoidable weighting in of the 5 Masai used. Admixture proportions are now basically the same. The pole was stretched and here it appears to be almost wholly dominated by the hidden West-African component.

In my previous "Chinese Mexican" experiment the "Chinese24" pole underestimated the actual admixture proportions and "Chinese5" grew correcting it. Here the larger Masai pole overestimates it, and "Masai5" shrinks to correct it.
Naturally this is easier with just two populations, since there's no extra-component to be "eaten-up".

Now back to North Africa
All Masai, all Yoruba and all Basque as 3 poles:
"Masai" to "Yoruba": 38
"Masai" to "Basque": 117
"Yoruba" to "Basque": 163
(distances are slightly different to those of "Yoruba24" since I used all Yoruba here and Yoruba are a poor substitute in this case I believe)

Only 5 Masai as the "Masai5" pole, remaining poles full-set:
"Masai" to "Yoruba": 82
"Masai" to "Basque": 99
"Yoruba" to "Basque": 182
Tunisians are similar to Northern Moroccans. These are the two large fertile areas and mostly agricultural; while other sampled populations live in regions with some fertile but other less fertile areas more suited to pastoralism.
Moroccan-Jews mostly lose their "Yoruba"-like component.
So Masai are likely not a good proxy for the component it's picking up here...
That's all I'll imply by now. Don't pay too close attention to fst since it's possible the pole's inadequacy made them eat up quite a bit of the other components. Masai have some Northern admixture making this more likely.

I don't have more time right now, so tomorrow I'll try to go further...
I should warn you I think in a very probability-based manner and all I need to jump is to feel a significantly higher probability in the validity of a trick/interpretation. I don't play the "wrong"/"certain" game. Good bets are valid even if sometimes "erroneous", because they can add up (as long as they don't bankrupt you). I won't tie my hands up for fear of being wrong.

I soon may have to review my early replacement from Egypt model for North Africa. Native Northwest Africans may also have had incipient Neolithic capabilities just as Northeast Africans... Maybe other Green Saharan melting pot populations shared them too. The Green Sahara may well have been an incipient Neolithic Eden. West Africans, East Africans, the Nile Core, and aboriginal Northwest Africans may all be the sons of another Neolithic Eve.

Also Dienekes comments on a new paper possibly relevant.

So take these experiments with a pinch of salt right now. It's only food for thought...


  1. When you use only the 5 Maasai, many of the low-frequency West Eurasian alleles that exist in the Maasai population become "fixed", and the Maasai appear more African than they are; correspondingly they appear more like Yoruba, and hence the African Americans' admixture proportions become more like those you get when you use Yoruba.

  2. Yes, but I've seen ADMIXTURE stretch totally unrelated poles as well to include hidden components such as Chukchi fishing out the West Eurasian and !kung the African elements in Middle Easterners. I've actually seen it do that repeatedly. I think some flexibility concerning possible variability in a postulated parent population, but not represented in pole individuals available, was written into the supervised mode's algorithm. I think it can be exploited to find hidden elements by using less individuals or totally unreasonable poles.