I have decided to generate a new major data dump of ADMIXTURE results. In comparison to previous such experiments:
- The focus is entirely on West Eurasians (Caucasoids).
- I have excluded all potential relatives from the source datasets, as well as several populations that tend to create uninformative clusters of their own (e.g., Druze or Ashkenazi Jews); exceptions are populations of great anthropological interest (e.g., Basques).
- I have included all relevant Dodecad Ancestry Project populations with 5+ participants.
- I have developed a new way of "framing" the region of interest by choosing appropriate sets of individuals from outside of it.
"Framing" populations
I have, since the beginning of my ADMIXTURE experiments, emphasized the importance of including appropriate population controls designed to squeeze out minor distant admixture in populations of interest, so that it does not confound the inference of region-specific components.
This leads to a problem: there are many possible sources of admixture. For example, we do not know a priori which set of African populations may have contributed to Caucasoid populations, or which set of East Asian ones. We could choose e.g., the Yoruba and the Chinese to represent Sub-Saharans and East Asians, but that might exclude possible sources of variation, and lead to Yoruba- and Chinese- specific clusters rather than more general Sub-Saharan and East Asian ones. If we included more population controls, we would cover more possible sources of variation, but ADMIXTURE would infer components of little interest (e.g., between Pygmies vs. Bushmen or Mongols vs. Chinese)
To avoid this, I propose to create meta-populations consisting of a single individual from many populations, i.e., a Yoruba, a Mandenka, a San, a Mbuti Pygmy, etc. for Sub-Saharan Africa, or a Miaozu, a Han, a Mongol, a She, a Hezhen, etc. for East Asia. That way we are both helping ADMIXTURE infer general components, while at the same time preventing it from inferring non-region specific ones.
Results
The entirety of the results presented here can be downloaded. They include:
- Population sources
- ADMIXTURE proportions for populations
- Fst divergences between components
- Population portraits showing individual level variation
See
spreadsheet and associated
bundle (or
here).
At K=3, we observe the emergence West Eurasian, Sub-Saharan, and East/South Asian components.
The impact of the Sub-Saharan component is felt most distinctly in North Africa and the Near East, especially among Arabs; the impact of the East/South Asian one in West Asia and Northeastern Europe, especially among Finnic and Turkic speakers.
It is interesting to note that 39.8% of the
Indian_D sample is assigned to the E/S Asian component. I had previously
estimated in a roundabout way, and in a slightly smaller sample that the Ancestral South Indian component in Project participants was 33.3%, so ADMIXTURE has roughly managed to infer correctly that about 1/3 of this Indian sample's ancestry is more closely related to East Asians than to West Eurasians.
At K=4, the first split within the Caucasoid group appears: a component centered onn Europe, and one on West/South Asia.
Many populations possess both these components in clinal proportions.
The European component shrinks to insignificance in Arabians, such as Saudis and Yemenese.
The West/South Asian component shrinks to insignificance in Northeast Europeans, such as Finns, Lithuanians, north Russians, and Chuvash.
At
K=5, a new
Mediterranean component emerges. This is highly represented in populations to the North, South, and East of the Mediterranean sea.
This component is noteworthy for its absence in India and Northeastern Europe.
In Northeastern Europe, the Mediterranean component is hardly represented at all, whereas the West/South Asian component, freed of its K=4 Mediterranean associations now makes its appearance.
Conversely, in the West Mediterranean, among Basques, Sardinians, Moroccans, and Mozabites the West/South Asian component vanishes to non-existence.
At
K=6, a
North African component emerges.
Notice its presence in the Near East and parts of Southern Europe.
The two regions can be contrasted in terms of their African components, with very high North/Sub-Saharan African ratio in Europe vs. much lower in the Near East.
The explanation for this seems straightforward, as Europe was affected by North Africa in prehistoric and historic times, whereas the Near East also shares a border with more southern parts of the African continent, as well as the potential influence of the medieval slave trade that seems to have affected Muslim Near Eastern populations disproportionately.
At
K=7, a
Southwest Asian component emerges which is highest in Arabia and East Africa. I could've called this Red Sea, but I've reserved this name for a similar component that emerges at higher K.
It is clear that this is the main Caucasoid component present in East Africa.
It vanishes to non-existence in the Northern fringe of Europe, in the British Isles, Scandinavia, and among the Finns and Lithuanians.
Another interesting aspect of its distribution is its presence in Pakistan but not India. Perhaps, in this case, it reflects historical contacts between the Islamic Near East and parts of South Asia.
At
K=8, we observe most of the familiar components from the K=10 analysis of the Dodecad Project. However, the use of the framing populations has meant that these components emerge before either Africans or East Eurasians split.
Now, the South Asian component appears, which swallows up most of the E/S Asian component that previously linked South with East Asians. This component extends a great way to the Near East and eastern parts of the Caucasus.
Quite interestingly, the remainder of the Caucasoid component in South Asia that is not absorbed by the new South Asian component seems to be split between the West Asian and North/Central European components, with an absence of the South European component.
It is among the Lezgins of the Caucasus that such a combination occurs, on the western shore of the Caspian Sea. The same combination of Caucasoid components also occurs in Uzbeks and Chuvash.
I conclude from this that the Caucasoids who entered South and Central Asia were probably derived from the eastern fringes of the Caucasoid world where only the West Asian (in the south) and North/Central European (in the north) are in existence. The area around the
Caspian Sea seems like an excellent candidate for their origin, as I have
speculated before, as that region has two important properties:
- It is transitional between predominantly N/C European populations to the north and predominantly W Asian populations to the south
- It is the border of the influence of the S European element, with Georgians possessing some of it, while Lezgins do not.
At
K=9, we see the emergence of specific
Sardinian and
Basque components. Normally this is undesirable, but, I believe this breakup serves to divide the previously inferred
South European component meaningfully.
What was South European in lower K seems to have an Atlantic vs. Mediterranean dimension, with the Basque/Sardinian ratio being particularly high in the Atlantic facade of Europe. Conversely, this ratio is low in the Mediterranean as we move eastwards: it is already low in Italy and the Balkans and becomes virtually zero in Cypriots, Armenians, and Levantine Arabs.
North Africa is also particularly interesting in having a low Basque/Sardinian rate, even in Morocco. It appears that Sardinians are a much better proxy of European influences in the region than Basques are.
K=10 is particularly exciting because, for the first time, there is clear evidence of structure in the
North/Central European component that can now be split, for the first time, into
Northwestern and
Northeastern ones.
The NW European component is maximized in Orcadians, and people from the British Isles in general, as well as in Scandinavia. These populations have a low NE/NW ratio, as do the French, Iberians, and Italians.
Conversely, Balto-Slavs have a high NE/NW ratio.
Interestingly, Greeks have a balanced NE/NW ratio (1.2), intermediate between Italians and Balto-Slavs. Similar balanced ratios are also found among Lezgins (1.08), Turks, and Iranians. I conclude that Slavic or other Eastern European admixture cannot account for the totality of presence of this component in Greeks.
Indians have a 1.8 NE/NW ratio. In Pakistan this is 6.5, in Uzbeks it is 2.9, and in the North Eurasian_Ra it is 14.2. My conclusion is that a single migration of steppe people from eastern Europe cannot account for the presence of North European-like genes in Asia.
I propose that a palimpsest of population movements has brought such elements into the interior of Asia: the migration of the early Indo-Iranians from West Asia or the Balkans with a balanced NE/NW ratio, and, the migration of steppe people from Eastern Europe with a high NE/NW ratio. The latter, did affect much of Asia, but it is in India, where Iranian groups did not penetrate in great numbers the lower ratio of the Indo-Aryans has been best preserved.
The case of the Finns is also interesting, as there is a surplus of NE over NW European elements. Their position is intermediate between Scandinavians and Lithuanians/Russians but toward the latter. So, Finns appear to (i) have a substratum similar to Balto-Slavs, (ii) to be influenced by Scandinavians, and (iii) with a balance of East Eurasian elements (5.8% at this analysis) preserving the legacy of their linguistic ancestors from the east. At present it is difficult to determine how much of the NE European component in Finns is due to their eastern ancestors who were presumably mixed Caucasoid/Mongoloid long before they arrived in the Baltic, and how much was absorbed in situ.
At
K=11 the
Ethiopian/East African component emerges, absorbing some of the Red Sea and Sub-Saharan components from the previous K=10 run.
In comparison to the East African component of the Dodecad Project analysis, this component is closer to West Eurasians than to Sub-Saharan Africans, and a residual Sub-Saharan element remains in the two East African (Ethiopian and East_African_D) population samples. Presumably this is due to the more complete sampling of Sub-Saharan genetic diversity using the Sub_Saharan_H "framing" population.
Outside Africa, both E African/Sub-Saharan components are present in the Near East and North Africa with higher E African/Sub-Saharan ratios in the Near East and lower ones in North Africa.
In Europe, there are low such ratios in the few populations where African admixture is present, together with some N African. We can probably conclude that African admixture is mostly due to North Africans, and African-influenced Near Eastern populations, rather than directly from Sub-Saharan Africa.
At K=12 the first uninformative cluster emerged, centered on Iraqi Jews, hence I decided to stop the analysis at this point.
Population Portraits
There is a plethora of population portraits in the download bundle, showing how admixture proportions vary in individuals within populations, and how they vary between successive K.
Here is, for example, the K=11 portrait of Cypriots. A picture of overall homogeneity of this sample emerges, but notice how the NW European and NE European have disjoint presence in the Cypriot individuals, with 5 having some of the former, 6 having some of the latter, and only 1 of these having both.
Compare with Lezgins (right) where these two components occur in all individuals. Whatever this admixture represents, it must be old enough if it is so uniformly distributed in the population.
Here are the Georgians at K=10. Notice that their NE European component is unevenly distributed, and in every case where it occurs it is accompanied by a thin slice of East Asian. This may well indicate partial Russian or other Eastern European ancestry in these individuals.
Side-by-side comparisons are also quite useful. Consider Armenians vs. Lezgins vs. Iranians at K=7
Notice how Lezgins, who live north of the Caucasus mountains possess some of the N/C European component, which the Armenians, who live to the south of them lack. This should come as no surprise, as the Lezgins inhabit parts of the ancient Sarmatia Asiatica. Compare with Iranians, who are differentiated by their Indo-European Armenian neighbors by the presence of a "S Asian" component, which, in turn, ties them to their Indo-Aryan linguistic relatives.
Much more can be said, but I'll let readers explore the data on their own, and draw their own conclusions from them.
