search this blog

Sunday, March 15, 2015

Modeling the ancestry of Yamnaya with qpAdm


I've been playing around with the new qpAdm program and the Haak et al. dataset over the past few days and managed to come up with what I think are some very promising results. For instance, the Yamnaya genomes from the Samara Valley and surrounds fit rather well as 0.514 Samara hunter-gatherer + 0.486 Georgian (std. errors 0.032, chisq 3.890).

This is an interesting outcome, mainly because Georgian is a Kartvelian language, and linguistics data suggest that the early Indo-Europeans - presumably the Yamnaya nomads or their ancestors - were in close contact with Proto-Kartvelian speakers. Moreover, even though the Yamnaya males tested to date all belong to Y-chromosome haplogroup R1b, which they probably inherited from their hunter-gatherer ancestors, because the Samara forager also belonged to this haplogroup, some of their mtDNA lineages appear to be derived from the Caucasus and/or nearby areas of the Near East.

However, the main problem with this analysis is that it's attempting to model an ancient population as a mixture of a modern one. Indeed, my estimate is that present-day Georgians harbor around 20% of the so called Ancient North Eurasian (ANE) component, which probably arrived in the Caucasus from the Eurasian steppe (see here). If so, then the qpAdm run might be overestimating the non-steppe admixture in the Yamnaya genomes by at least 10%. Nevertheless, I'm quite happy with this result as I await ancient DNA from the Caucasus and Near East.

By the way, I also pretty much nailed the Corded Ware sample: 0.73 Yamnaya + 0.27 Esperstedt_MN (std. errors 0.060, chisq 2.621). Admittedly, an identical result for the same genomes was reported months ago at the ASHG 2014 conference (see here), but that's OK, because it means I'm on the right track.

qpAdm is easy to run, but the quality of its output heavily reliant on the outgroup or "right set" of populations picked by the user. As far as I can see, the following ten populations (a subset of the "magic set" of 15 from Haak et al.) produce the most robust outcomes when analyses are limited to West Eurasian groups.

Biaka
Bougainville
Chukchi
Eskimo
Han
Ju_hoan_North
Karitiana
Mbuti
Ulchi
Yoruba


Why do they work so well? I really have no idea, but through simple trial and error I found that some of the others from the "magic set", in particular the Ami, produced much poorer results.

I'll probably end up posting a whole catalog of qpAdm output in the comments section below over the next couple of weeks. I'm open to suggestions about the models to test and how to improve my runs.

Citation...

Haak et al., Massive migration from the steppe was a source for Indo-European languages in Europe, Nature, Advance online publication, doi:10.1038/nature14317

Friday, March 13, 2015

Yamnaya-related ancestry proportions in Europe and west Asia


Here's a quick and dirty attempt to flush out a Yamnaya-specific ancestral component with the ADMIXTURE software and a few Yamnaya genomes from the recent Haak et al. paper: K6 spreadsheet.


Obviously, we'll need many more ancient samples from the vast Yamnaya horizon to be able to estimate direct Yamnaya ancestry in modern populations with any great confidence. But I'd say this looks like a very reasonable attempt, with more or less comparable results to those published by Haak et al. (see here).

Please note that this wasn't a supervised run. In other words, I didn't mark the Yamnaya genomes as reference samples with the aim of creating a cluster from them.

However, I initially excluded all individuals from northeastern Europe, the north Caucasus and South Asia from the analysis. The reason I did this was because samples from these regions have a peculiar habit of creating very robust clusters in ADMIXTURE, which is useful when looking at recent variation and wanting low cross validation errors, but not so great when trying to resurrect genetic components from the depths of prehistory.

Once I had a dataset that was forcing the algorithm to focus its attention on the ancient genomes and producing consistent results, I tested the problem samples in batches of 5-10, thus making sure they didn't skew the analysis.

Interestingly, the Yamnaya-specific component peaks in Udmurts, who live close to where the Yamnaya samples were collected. This can hardly be a coincidence.

In any case, I'm hoping to look at this issue in more detail soon with the help of qpAdm, a new program released recently with the updated ADMIXTOOLS package (see here). Based on f4 statistics, qpAdm is specifically designed for analyzing ancient admixture events.

Citation...

Haak et al., Massive migration from the steppe was a source for Indo-European languages in Europe, Nature, Advance online publication, doi:10.1038/nature14317

Wednesday, March 11, 2015

New release of ADMIXTOOLS with two additional programs


ADMIXTOOLS 3.0 is now available at github via the Reich Lab site. The updated package includes minor bugfixes and improvements and two new programs: qpWave and qpAdm for studying migration and admixture. Scroll down the page at the link below.

Reich Lab software

Documentation is minimal, but I'm told that users of the old ADMIXTOOLS should be able to get things running. I haven't had a chance to check it out yet, but I'm looking forward to trying qpWave and qpAdm, hopefully this weekend.

Update 28/06/2020: Major updates to ADMIXTOOLS