Somali qpAdm models using new ancient genomes

So, I asked David over at Eurogenes to run Somalis as a mixture between South Sudanese people and Natufians in order to see how well the model would fit using a formal statistical method like qpAdm and he got some pretty surprising results overall:


Natufian + Sudanese (south):

Sudanese: 54%
Natufian: 46%

Neolithic Levant + Sudanese (south):

Sudanese: 54%
Neolithic Levant: 46%

Neolithic Levant + Chalcolithic Iran + Sudanese (south):

Sudanese: 55%
Neolithic Levant:  34%
Chalcolithic Iran: 11%


Now, what's going to surprise you is that the third model is the one that fits the best, and by a long shot when compared to the first model. Natufian + Sudanese (south) fits the worst (chisq: 26.256, tail prob: 0.09%, std. errors: 0.009), Neolithic Levant + Sudanese (south) fits much better (chisq: 7.593, tail prob: 47%, std. errors: 0.006) and Neolithic Levant + Chalcolithic Iran + Sudanese (south) fits even better (chisq: 4.975, tail prob: 66%, std. errors: 0.057).

The last one almost fits as well as a Corded Ware sample being modeled as ~70% Yamnaya & ~30% Esperstedt Middle-Neolithic (chisq: 2.621, std. errors: 0.060) which roughly fits with the data from peer-reviewed studies like Haak et al. 2015:

This oddly reminds me of some models the new Lazaridis et al. pre-print shared where they were asserting that Somalis were a mixture between Mota & population along the Iran_ChL→Levant_BA cline:

I didn't make much of the above at the time. For one, Mota is a poorer fit for Somalis' African ancestry than the South Sudanese (due to various reasons alluded to here), and it made little sense that Somalis' West Eurasian ancestry corresponded better with Bronze Age Levantines and Copper Age Iranians than Neolithic Levantines, for example. At least in my humble opinion.

I figured Lazaridis & company just didn't try a different model that would probably fit better but it's now odd that this fits a bit well with what the above qpAdm models imply which is that "Neolithic Levantine + Chalcolithic Iranian + Sudanese (south)" fits much better than "Natufian + Sudanese (south)" and somewhat better than "Neolithic Levant + Sudanese (south)".

Future analyses and data will be needed but I should point out that some current ADMIXTURE runs don't seem entirely supportive of such a model but ADMIXTURE is not necessarily as precise as qpAdm can be.

For one, qpAdm is preferable because it outright allows you to take a Natufian and then a Neolithic Levantine and see which one you have a greater affinity for (mixture wise) but ADMIXTURE is a lot more messy in that it allows all of these clusters to form among various modern & pre-historic populations and could thus be more prone to producing perhaps more clunky results. Formal statistical methods like qpAdm also seem to be "drift resistant" / resistant to being skewed by recent genetic drift and can thus notice deeper ancestry better than ADMIXTURE to a certain degree.

But we should see what some other analyses say like d-stats and tree-mix. I'm skeptical about the third model in particular (for the time being), despite how well it fits.

Reference List:

1. The genetic structure of the world's first farmers, Lazaridis et al. 2016

Notes: 

1. Link to the full qpAdm results.