Detailed specific formal study of IA populations affecting Iranics

[Zoro]

Background:

About a year ago I decided to commission a serious detailed and specific study of Iron Age and Bronze Age populations affecting Iranics using formal methods. I was tired of amateurish calculators giving results that many times didn't make much sense.

I commissioned Eurasian DNA to do this study because of their proficiency in DNA processing and innovation in adapting formal methods to specific purposes and because they do more customized studies on West, Central, and South Asia than other outfits.

One of the first things Dilawer from Eurasian DNA told me is that if I wanted to get as accurate as possible results I would have to whole genome sequence myself since the 1240K panels ancients are genotyped have low overlap with 23andme, FTDNA, and AncestryDNA panels. Also he mentioned that he has noticed some genotyping errors with the commercial companies when he has compared their samples with the same sampled done with 30X sequencing.

So I decided to do whole genome sequencing for me and my mom so that Dilawer could use our data with the 1240K ancients for maximum accuracy and SNP overlap.

The other thing Dilawer cautioned was that there was a limited number of modern samples and populations that had maximum overlap with the 1240K SNP ancients. I told him I was fine with that since I just wanted Kurds compared with a couple of other West Asians and maybe a couple of E Europeans and S Asians.

[Zoro]

Methods Used:

Dilawer adapted the ADMIXTOOLS qpWave method using the following 11 outgroups to get rid of junk SNPs that were shared by many due to very deep ancestry:

South_Africa_2000BP.SG
Russia_Ust_Ishim.DG
Russia_Shamanka_Eneolithic
Onge_1000G
Russia_HG_Karelia.SG
Serbia_Mesolithic_IronGates
DevilsCave_N.SG
Anatolia_N
Karitiana.DG
Iran_GanjDareh_N
Georgia_Kotias.SG

So in other words these outgroups would help base the analysis on derived alleles and not shared ancestral alleles which skew alot of calculators.

Dilawer is also very strict with the quality of ancient samples so in instances where there are many samples of a population he only used the highest quality ones and discarded the others.

Here are the quantities of samples from each outgroup population (n):

2 South_Africa_2000BP.SG 3
3 Russia_Ust_Ishim.DG 1
4 Russia_Shamanka_Eneolithic 10
5 Onge_1000G 6
6 Russia_HG_Karelia.SG 1
7 Serbia_Mesolithic_IronGates 29
8 DevilsCave_N.SG 4
9 Anatolia_N 29
10 Karitiana.DG 3
11 Iran_GanjDareh_N 4
12 Georgia_Kotias.SG 1

[Zoro]

Formal methods used

I found this regarding the methods Dilawer used https://bodkan.net/admixr/articles/tutorial.html

To run qpWave, you must provide a list of left and right populations (using the terminology of Haak et al. 2015 above). The aim of the method is to get an idea about the number of migration waves from right to left (with no back-migration from left to right!). This is done by estimating the rank of a matrix of all possible F4 statistics

F4(left1,lefti;right1,rightj),

where left1 and right1 are some fixed populations and the i and j indices run over all other possible choices of populations.

The qpWave() function returns a data frame which shows the results of a series of matrix rank tests. The rank column is the matrix rank tested, df, chisq and tail give the degrees of freedom, X2 value and p-value for the comparison with the saturated model (the p-value then indicates which matrix rank is consistent with the data - see example below), and dfdiff, chisqdiff and taildiff give the same, but always comparing a model to the model with one rank less.

[Zoro]

Targeted Ancients:

POPULATION

Turkmenistan_IA ; n=1

Sarmatian-Kazakhstan ; n=2

Kura-Araxes-EBA ; n=4

Iran_C_HajjiFiruz ; n=4

Iran_IA_Hasanlu ; n=4

Iran-BA-HajjiFiruz ; n=1

Turkmenistan_Gonur_BA1 ; n=8

Alan-Russia : n=5

Saka-TianShan-IA : n=6

Iran-IA-HajjiFiruz : n=1

Iran-ShahrISokhta-BA1 : n=5

Sarmatian-Russia : n=6

Pakistan-Barikot-IA : n=3

Sintashta-MLBA ; n=16

[Voskos]

Cool.Where's the results?

[Zoro]

Cool.Where's the results?

I was waiting for some further explanation from Dilawer on the technical aspect of the methodology because it's difficult to understand so that I can explain it better. I'll start posting charts as soon as he gets back to me.

[Zoro]

I think I understand the methodology now. First this is what he said

With a sufficient number of outgroups which outnumber sources (n) the qpWave matrix has maximum n-1 independent columns (= rank) and minimum zero nontrivial column (i.e. zero matrix). For example, two unadmixed West Asian left populations are a sister group to each other against all non-West Asian right populations. Then, any ��4(WAsian1, WAsian2;EAsian1,EAsian2)will be zero.

As per Dilawer he is adapting the rank=0 matrix in qpWave which would have a passing p-value for 2 source populations which are very related to each other to infer degree of relatedness for any 2 source populations.

As per Dilawer if 2 sources are symmetrically related to ALL the outgroups the p-value would be high and X2 would be close to 0.

If they are very differentially related to ALL the OUTGROUPS Chi-squared (X2) will be very high

[Zoro]

Only they most densely genotyped samples were used in the study to get to 700,000 overlapping SNPs with the ancients. That's why there is a limited number of moderns used.

I'll start with Chalcolithic Iranian Zagrosian HajiFiruz from the Kurdistan area. Looking at the results it looks like there is no modern population in the area that would be extremely similar because of the Indo-Europeanization of the area after his time.

1- Haji-Firuz-CHl immediate relatives if they still existed would score 1400 in this chart. Even though Kurmanji Kurds and Armenians are close they are sufficiently different due to IE of the region after his time.

2- Notice the distance between Estonians and Indo-Iranians is very large at this time. You will see that this distance decreases with the Iron Age HajiFiruz after there is influx of steppe into this region.

[IMG][/IMG]

[Zoro]

Here is Iron Age Hasanlu. Here you see Iranics are getting much closer to E Europeans after Indo-Europeanization (look at the scale 600 instead of 1400)

The maximum Chi-sq spread between Iranics and Estonians is only about 550 vs about 1200 in the Chalcolithic period before Iran had influx of steppe

[IMG][/IMG]

[Zoro]

With respect to the relatedness to the higher quality 2 Sarmatian Kazakhstan samples N Ossetians and Kurmanji Kurds top the list and are shoulder to shoulder. However you'll see later that the gap between N Ossetians and Kurmanji Kurds grows a little with the 6 Alan samples. N Ossetians take the lead by a little.

I was surprised to see Armenians that far down.

[IMG][/IMG]