I ran supervised admixture on the Metspalu et al dataset using my reference 3 data. Here's the spreadsheet for Metspalu admixture results. You can compare with the reference 3 results.
Here's our bar chart for Metspalu results. Remember you can click on the legend or the table headers to sort.
These are very different from Dienekes for some reason.
UPDATE (Dec 13 10:04am): I found a major error. I had used the population info file I had downloaded from the paper instead of my reformatted one and thus I had not merged that info with the correct IDs with the admixture results. So the previously posted results were junk. I have fixed that now and the results are as expected.
Zack, while the Reference 3. K=11 ADMIXTURE runs were fairly informative for South-Asians, in the sense that it inferred a predominantly West-Eurasian, South Asian-specific component; an ASI proxy (Onge) and two exogenous West-Eurasian elements specific to the Near East and Europe, an an East Asian serving as a proxy for SE Asian influence in SA, I would certainly not take it as the final word on the admixture variation of South-Asian populations (I'm not accusing you of taking it as the final word, of course). The lack of a West-Asian/trans-Caucasian centered component, which has been shown as the source for the vast majority of the West-Eurasian admixture in India/Pakistan, is what perturbs me most. In light of this, the results look very, very off.
- The Tamil Nadu Brahmin average doesn't corroborate with the average K=11 percentages for the Tamil Brahmin participants of your project. The Onge is usually around 25% (as opposed to 31%). The European and SW Asian also typically vary on the levels of 7-12% and 7-9% respectively.
- The Uttar Pradesh Brahmins in thus run are oddly more Onge and less European/SW Asian than some of the peasant castes, such as the Chamars, Kanjar, and notably, the Kurmi agriculturalists. This neither corroborates with the consistent signal seen across genome blogosphere, that, outside of Northwest India, the Brahmins differ appreciably from the generic populace of the same geographic area; nor does it agree with the actual ADMIXTURE results from both the Metspalu et al. study and Dienekes' analysis. In fact;
The Dodecad K12a admixture proportions for the North-Central Indian participants and the U.P references only supports the above. Please see compiled, tabulated results in the last part of this post @ DNA-Forums.org - http://dna-forums.org/index.php?/topic/13354-dodecad-ancestry-project/page__view__findpost__p__293154
I have fixed the error and the admixture results are as expected now.
As for being "final word", reference 3 admixture isn't it and neither will be the new admixture results I am working on now.
If I understand correctly, you estimated the admixture proportions of the Metspalu et al. (2011) samples using the ancestral components derived from your reference 3 dataset. This can be problematic for populations that are at the edge of variation (e.g., South Indians). The following post explains what may be going on:
http://dienekes.blogspot.com/2011/10/further-caution-on-admixture-estimates.html
Yes, that can be a problem, but wasn't in this case since I have a fair number of samples from India in any case. The Metspalu samples are not that different from what I have before.
I calculated ANI from the spreadsheet here as well as from the data posted by Dienekes.
https://docs.google.com/spreadsheet/ccc?key=0ArJDEoCgzRKedGdRbkxKMDdlZkJWc21tdkpldWxwVmc#gid=0
............................................Zack_ANI...Dienekies_ANI...Caste
Bengali........................................52%.........49%..............n/a
Bhunjia.......................................39%.........13%...............6
Brahmins Tamil Nadu...............,....57%.........58%..............1
Brahmins Uttaranchal....................51%.........43%..............1
Brahmins Uttar Pradesh................63%..........62%..............1
Chamar.......................................46%..........39%..............5
Chenchus....................................44%..........39%..............6
Dharkars.....................................54%..........52%..............5
Dhurwa......................................40%...........13%..............6
Dusadh.......................................47%..........46%..............5
Gond..........................................46%...........34%.............6
Hakkipikki...................................44%...........37%.............6
Kanjars.......................................53%...........50%.............6
Kol..............................................48%..........44%..............6
Kshatriya.....................................59%,,,,,,,,,,60%..............2
Kurmi..........................................54%..........52%.............4
Kurumba.....................................48%..........46%.............4
Lambadi......................................54%..........52%.............6
Meena.........................................60%..........64%.............2
Meghawal....................................53%..........52%.............5
Muslim........................................54%...........51%............n/a
Naga...........................................32%............1%.............6
Nihali...........................................41%..........24%............6
Piramalai Kallars...........................46%..........44%............4
Pulliyar.........................................42%..........23%............6
Tamil Nadu Scheduled Caste..........49%..........48%............6
Tharus.........................................55%..........54%............6
Uttar Pradesh Scheduled Caste........50%...........47%.........5
Velamas.........................................49%...........50%.........4
Brahmin = 1
Kshatriya = 2
Shudra = 4
Scheduled Caste/Dalit = 5
Tribe = 6
The ANI values generally agree for ANI greater than about 40%. But for values less than 40%, I think the Dienekes data are more to be trusted.
The Uttaranchal Brahmins with an ANI (Dienekes) of 43% are an outlier because of East Asian admixture.
The Meghawal of Metspalu et al. are quite different from that of Reich et al. The more recent data show an ANI of 52% which is more consistent with the Dalit status of the Meghawal. But Reich et al. reported 60%. Perhaps Reich et al. mislabeled their samples.
Metspalu et al. data show an ANI of 48% for Tamil Nadu scheduled castes. This looks too high to me. It should be more like 40%. Either the samples are mislabeled or the people have misrepresented themeleves as scheduled caste to gain the benefits of reservations in government jobs and educational institutions.
What is the formula you are using to infer total West-Eurasian percentages (for K12a)?
Also, any idea what the ethnic backgrounds of the following participants in the Dodecad project are? They look to be South Asian, so..
DOD201
DOD583
DOD782
DOD078
DOD449
DOD666
DOD720
DOD698
DOD757
DOD749
DOD414
DOD753
DOD430
DOD819
DOD782 is the same as HRP0155. A Sinhalese Govigama from Sri Lanka. Me!
Ah, yes, thank you. I was already aware that DOD782 was you, though. Sorry I listed you under the unknown, potential South-Asians.
ANI=Mediterranean + North_European + Caucusus + Gedrosia + 0.16*South_Asian
Thanks a lot, Balaji. Balaji, would you have any idea of the backgrounds of the IDs of the Dodecad participants I listed? Are you participating in the project yourself, out of curiosity? Thanks in advance.
How much indian am i really (ANI-ASI)? I think its not the same as my Onge+Southasian (which would be 25-30%) because than the reall indians here would be 100% combined, but they are also not, and people with 4 indian grandparents score differently eatch. I never saw on this projects somebody with 100% southasian component, highest i saw was 70-80%.
It's complicated by the fact that your European ancestry would also result in you having some ANI (and South Asian component in this run). You ASI estimate is about 14%. Assuming your long ago ancestors belonged to North India/Pakistan, you are probably 25-50% "Indian". This is just a back-of-the-envelop calculation. I might have a better estimate once I can better distinguish between North Indian/ANI and European/Caucasian/West Asian etc.
thanks zack. 50% would make sense, since im 60% roma gypsy.
The question remains whether Reich's ANI is even a legitimate ancestral component? I was under the impression that ANI was simply a convenient mnemonic that was a catch-all term for the total inferred West-Eurasian component in Pakistan/India, and that the so-called ANI was a composite and what this composite comprised of varied from region and ethnic group. The vast majority of it being closest to trans-Caucasus populations and West-Asians, along with a secondary North-Eastern European component that is found in select groups and geographic areas. But looking at the latest U.P data, it looks to be widespread (in small amounts) as well.
We have also seen a Balochistan-centered component popping up at higher Ks. Dienekes' K12a Gedrosia component is the most recent one. In Harappa's initial K=12, we had a Brahui/Georgian twin peak. Although there is the possibility that this Balochistan-centered component has formed due to the effects of a few alleles reaching fixation or extinction in heavily isolated and drifted South-Central Asian groups.
This was meant to be a reply to Zack's comment, sorry - http://www.harappadna.org/2011/12/metspalu-ref3-admixture-results/#comment-12242
Sorry, sloppy language in my comment to which you are responding. Yes, I agree that ANI is probably not a single ancestry but a number of West-Eurasian-like ancestries bunched together.
My point to HRP15 was about being able to separate his European (Serb) ancestry from whatever West-Eurasian-like Indian ancestry he has. Since ASI varies so much over South Asia, it's not possible to use just that to estimate his quantum of Indian ancestry.
Didn't see this at all. Thanks for your answer, Zack.
As a quick aside, it does seem odd to me that the labelled Tamil Nadu Scheduled Caste reference has an inferred ANI score (calculated by Balaji) rather similar to the Tamil peasant castes (such as the Vellalar, Nadar and Vishwakarma participants), yet, the Pullayar who are Tamil out-castes or Harijans are substantially less West-Eurasian, at least as per Dienekes' K12a. Perhaps scheduled caste doesn't really entail traditional ethnic status per se, but rather an amalgamation of random samples that receives, at least in theory, benefits from the government of India due to their historically backward state.
What do you think of all this, Zack? Is linear regression based on ADMIXTURE runs a close-to-accurate way to infer total ANI/West-Eurasian scores? Or should we take these estimates with a pinch of salt?
The linear regression method has some serious flaws which is why I am not using it any more.
Also, my understanding is that Scheduled Caste can include quite a lot of different ethnic groups. They should generally be closer to the Harijans but it is possible to have significant variation there. Without knowing which specific group was sampled we can't really be sure.
Meena are not Kshatriyas. They are Schedule tribe.