With respect to the Haak et al. 2015 and after my own pretests using qpAdm I ran a serie of standard admixture tests in supervised mode to find out ancient admixtures among Europeans. Admixture gives a better way than qpAdm to use larger data sets, but without any quality indicators needs pretests if you want to run in supervised mode. I also used some present-day populations to obtain a better total fit in cases previous tests and studies gave reason to expect such being reasonable. In general my results are similar with those published by Haak. K-group sizes differ, in some cases my results look more plausibe than corresponding results in the above mentioned study, for example my admixture shows less Siberian admixture in West Europe, but also more Near Eastern admixture in South Europe. Those two admixtures came probably later to Europe which can be seen in their smaller expansion areas.
The first analysis (I'll do later more similar analyses) shows following k-groups:
1 WHG - Western Hunter-Gatherers
2 ENF . Early Neolithic Farmer (in Europe)
3 Yamnaya - samples from Yamnaya culture
4 East_Asia - Chinese She population
5 NA - Native Americans (Karitianas from South America)
6 SSA - Sub-Saharan African (Yoruba)
7 Near_East - Bedouins (BedouinB from Haak et al.)
8 Siberian - Nganassans
Sorted by ENF group averages. Click
here to see a large picture.
It's weird that populations such as yemenite jews are showing such consistent and high yamnaya component percentages. We should remember that without ancient populations, all we can know that the components are shared autosomal ancestry and we cannot know the direction of ancient migrations or indeed geographic origin of the components.
ReplyDeleteBut, I find your insistence on the absence of middle eastern ancestry in finns very weird. The fact is that we have now many samples of ancient european hunter gatherers, and they are all autosomally different from modern europeans, even Finns.
What could be responsible for this difference but middle eastern admixture? In case of finns it has to be ancient, because there is no history of modern admixture.
Yeah, I have pondered this same question and also made some tests. The answer is Yamnaya itself. It is a mixture of Iranian or Armenian like ancient people and Eastern hunter-gatherers. The Iranian/Armenian component gives the Yamnaya proportion for Near Easterners who have later genetic drift or admixtures differing from other sample groups used in my test. After replacing Yamnaya by Iranians or Armenians I get similar results with a bit different proportion sizes. It is hard to say which one (Iranian or Armenian) is best, because we speak about a mixture of modern and ancient people and modern people are more mixed with each other. Ancient Iranians (and Armenians) were different than today. This is something to think more and we need ancient Near Eastern samples from several places to find the answer why modern Near Easterners don't fit perfectly with now available ancient samples.
ReplyDeleteActually I don't insist any absence of Middle Eastern admixture in Finland, I only make a difference between ancient Middle Easterners and modern Middle Easterners. When I place ENF (Early Neolithic farmers in Europe) and modern unmixed Near Easterners into same admixture analysis I get two distinct groups and only ENF is present in Finland. This same observation was done by Haak in his recent study. You can see it on the page 214. So I don't deny the ancient Middle Eastern admixture in Finland and Western Europe. I do object seeing modern and 7000 years old Middle Easterners as fully similar people. If we can do something to see more why we don't do it and make the difference between old and young admixtures in Europe.
Modern and ancient middle eastern people are definitely different, but I think it could be fair to say that the farmers who entered europe were a genetic subster or genetically drifted group of ancient middle easterners. Similarly modern middle easterners are a genetically drifted (but in other way) group of ancient middle easterners, but of course they also have ANE and some have african ancestry).
ReplyDeleteBasically ENF is a closely related component to the Near_East component you found. But it is also local to europe as the people who brought it to europe, left the middle east 7000 years ago. They were not so different from modern middle easterners as some group separated by a continent.
I think you'll agree with me on this.
Early Neolithic farmers in Europecould have been similar to Near Easterners, but this is still speculation. I am a bit allergic for ethnic speculations, for some reason.
DeleteActually ENF seem to have been "ab origine" a 50-50 mix of (1) West Asian (+ minor African-like "Basal Eurasian") and (2) Paleo-Europeans from the Balcans. They do tend to West Asia by the side of Palestine or Arabia Peninsula but they are not at all the same. Spaghetti is clearly wrong.
DeleteWell, I'm pretty sure I'm not wrong, here's why: If Mr. Myllylä ran a supervised test with La-Brana, Loschbour, KO1, and other mesolithic and neolithic european huntergatherer samples, together with modern populations (this is in supervised mode, so trying to force the program to give the hunter-gatherers 100% of the WHG component) then that component would show all and no more of the ancestry shared with the hunter-gatherers, i.e., the entire WHG ancestry in europeans.
DeleteThe ENF has to be something else.
Unfortunately almost all ancient genomes are low quality and it is not possible to run Fst-distances against modern genomes, but I could try to make an indirect Fst-analysis. We can't compare Fst-values of ancient genomes and modern ones, but comparing for example ENF to several modern genomes, like West Europeans, Sardinians, Basques, Armenians, Iranians, Palestinians, Bedouins, can give us reasonable comparisons.
DeleteActually all groups of people can be said to be distinct, even neighboring ones like German and French, or west finns and east finns. 100% of the time, you can tell if a sample is from a french person or german person. It is quantity of differentiation or drift that matters. Is ENF so highly drifted from modern middle easterners? I don't think so.
ReplyDeleteWe'll see after we have ancient genomes from Near East.
DeleteOne correction more. I have no Jews on my blog entry and you are writing about my ad hoc test regarding Jews on ABF (Anthropological forum). It was meant to be only a demonstration about recent Near Eastern admixture and I mentioned that it is only an approximation. So it would be better to leave comments there, not here.
ReplyDeleteSorry, the page was 124 (Haak et al. 2015).
Sorry. I have not registered on ABF, but I might in future. I was referring to the ADMIXTURE graph in this post though.
DeleteOh, sorry. I thought you put the same graph here. Anyway, the ENF vs near_east question is interesting by itself. Maybe a future post?
DeleteYeah, we know that Near Easterners moved to Mediterranean Europe after the Neolithic era, so it is plausible that people in Mediterranean area have more Near Eastern admixture than for example Central Europeans. If you look the appearance of Central Europeans and South Europeans, you see that Central Europeans look less Near Eastern, although they have over 50% ENF according to Haak.
DeleteBut Yamna-related migrations also brought "Near Eastern" affinities with them. Hence my insistence on using two and not just one "Near Eastern" controls: Anatolian and Palestinian.
DeleteWhere does the "Near Eastern" reference come from (I'm assuming you're using whole populations rather than pre-filtered components)? My impression is that this "Near Eastern" population is one that is rather akin to Sardinians/ENF(Levantine or Cyprus probably) and not so much to Yamna (not Turks, Kurds, Armenians or Iranians, let alone the Caucasus).
I ran two analyses, replacing Yamnayas by Armenians and Turks. Both of those two are closer South Europeans than Yamnaya subtituting some ENF. I could run one analysis more with Armenian and Palestinians, replacing Bedouins by Palestinians. I expect that this could decrease WHG and other northern stuff i the south by one or another. But the basic quiestion still remains and it is "which kind of total sample set represents best modern people". We can play with admixture groups and populations without knowing how far we are from the optimal mix. For that reason I appreciate studies like Haak et al.
DeleteI didn't make any preselection, all populations are straight from the study data.
Thanks for the info. In my understanding the problem is that no single modern set represents best ancient West Asians but that we need (at least two): one representing the Highlanders and another the Lowlanders, one representing the Northerner West Asian (Jarmo-like) component that makes like 25% of Yamna ancestry and another one representing the Southerner West Asian (Natuf-like) component that makes like 50% (?) of ENF ancestry. Either both or none: there's no single West Asian tendency or "paleo-population" that represents well both.
DeleteIs this a free run or a supervised one? It's not clear in the exposition. If free run, how do you know what is EEF, Yamna, etc.?
ReplyDeleteSomething that is quite apparent in any case in your exercise is that it is heavily loaded with North European samples. If the run is supervised this should not be important, as components are pre-set, but if the run is free it will affect the results by overemphasizing Northern European components and minimizing southern ones.
If it is supervised, which are the reference samples for each component? In a few cases it may be obvious (WHG, ENF) but others not so much. In the case of Yamnaya, what are you considering as such: the whole Yamnaya genomic pool or only the non-ENF part? Etc.
Yes, this was run in supervised mode and the idea was to follow Haak et
ReplyDeleteal. 2015, which made good work and obtained high quality by seeking for lowest error rate (look page 124 and similar results). The main problem with Admixture is that in unsupervised mode it is very easily beaten by drifted populations. In supervised mode the problem is obvious - are you able to select right populations. But in this exercise I leant myself on Haak's quality checked results to obtain better results also using Admixture in supervised mode. This is not, however, foolproof because also supervised mode uses all populations, not only selected, although selected ones have higher priority. So the problem of drifted population still remains, although in lesser extent. I think that this could be the reason for overemphasizing something. So if this happened I
have not succeeded the avoid normal drawbacks of Admixture. Unfortunately. But the goal was expressly to decrease it by following Haak's selection and use supervised mode.
ENF is represented by Hungarian NE-samples, mainly by NE1. WHG is Loschbour. I used for Yamnaya the whole pool. Because Yamnaya is itself mixed, using it leads to certain uncertainty when used in testing wide range of populations, for example Middle Easterners who own a part of Yamnaya admixture. The Yamnaya result in this test can represent Near East or EHG instead of original Yamnaya heritage. Unfortunately this drawback is hard to avoid, but I can defend by stating that also Haak used Yamnaya :)
That's fine, as far as I can discern. Supervised mode should work well enough and avoid the sampling strategy problem. I'm wondering if who represents West Asians, because that's maybe critical, as West Asian - European interactions are the most intricate of all and West Asians are not at all homogeneous. I'd use two different populations: Turks (or Armenians or Georgians maybe but these last are a bit extreme) and Palestinians. But I do approve using whole Yamna genome instead of part.
DeleteBefore publishing this blog entry I made some Admixture tests using Armenians and Eastern hunter-gatherers. Eastern HG was added to represent ANE, but maybe some West Asian groups would work fine without EHG and give reasonable ANE. Those analyses confirmed Yamnayas being admixed. I could publish those results. There was some changes, but nothing tremendous. These Admixture runs are very quick and easy and the biggest problem is to decide what to do.
DeleteI'm not really interested in ANE, which is a LGM reference and not an Holocene one: there are differences between North or Highland West Asians and South or Lowland ones that are unrelated to ANE and should be discerned though. That's why I'd suggest Turkish and Palestinian controls (could be other populations but these two should perform well).
DeletePS- Yamna for example has Highland West Asian admixture, while ENF has rather something more similar to Palestinian admixture instead. Hence if you use only one West Asian reference, there may be confounding factors and this I suspect because Sardinians show too much West Asian for what is usual (typically they are almost exactly like ENF).
DeleteI'll run some alternative admixture analyses, but I can't say how big is the ANE impact in Northern and Eastern Europe, so new analyses will be interesting and need verbal analysis too.
DeleteNow that there are Bronce Age Armenian genomes available it would be great to see something like this but instead of Yamnaya using BA Armenian in one cluster and EHG (Karelia, Samara) in another one. I just wish those Armenian samples were some 1500 years older, though.
ReplyDeleteYeah, I have those Bronze Age samples and doing tests. Unfortunately I am only a couple days at home during this week and the next blog entry will be written likely at the end of next week. Armenian samples are surely interesting, as also new samples from East Europe and Scandinavia. As you wrote Yamnaya is likely an admixture and it would be interesting to find out which ancient mixture fits best to Yamnaya.
Delete