PDA

View Full Version : Playing with Iranian populations using NMonte Runner



BalkanKiwi
06-15-2020, 10:49 PM
On various calculators I've seen small amounts of Iranian populations appear, so now exams have finished I thought I'd do some quick modelling on GenoPlot to see how certain populations fit out of interest. I don't know the region well enough to know how they might fit in with my known ancestry (Ashkenazi?), but I thought I'd share regardless. If I use Vahaduo and go FreeForm using specific samples, the Seyyed % jumps a bit. I haven't checked other Ashkenazi members to see if its the norm or not. If any of our members from the region have some thoughts, you're more than welcome to share. If its anything real, it certainly a very long way back.

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.2493
Results:Irish71
Croatian13.5
English Cornwall12
Iranian Seyyed3.5
Vahaduo

Distance: 1.5892% / 0.01589188
52.4 Irish
30.0 English_Cornwall
13.2 Croatian
4.4 Iranian_Seyyed

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.2545
Results:Irish69
Croatian14
English Cornwall13.5
Iranian Lor3.5

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.2609
Results:Irish70
Croatian13.5
English Cornwall13
Iranian Fars3.5

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.2642
Results:Irish71
Croatian14.5
English Cornwall11
Azeri3.5

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.2899
Results:Irish66.5
English Cornwall17
Croatian13.5
Iranian Zoroastrian3

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.2764
Results:Irish69.5
Croatian15.5
English Cornwall12.5
Iranian Mazandarani2.5

digital_noise
06-16-2020, 01:11 AM
Can you post the datasheet you are using? I'll run my daughter, she's half Iranian

BalkanKiwi
06-16-2020, 01:39 AM
Can you post the datasheet you are using? I'll run my daughter, she's half Iranian

I can give you two of the Iranian sample coordinates I use in combination with the Croatian, Irish and English ones, unless you want the coordinates for the 3 of those as well.

Iranian_Lor:LORII48,0.099026,0.108662,-0.058077,-0.029393,-0.048317,-0.01004,0.002585,-0.001385,-0.028838,-0.006014,-0.001624,0.001798,0.001041,0,0.013301,0.009944,-0.003781,0.007728,0.006788,-0.01063,0.010981,-0.008532,-0.007148,-0.013496,0.004311

Iranian_Seyyed:SHII51,0.078538,0.080227,-0.058831,-0.027455,-0.033545,-0.007809,0.001175,-0.011538,-0.015339,-0.015126,0.003085,-0.004196,0.009068,0.00289,0.010179,0.006364,-0.011474,-0.001267,0.000754,-0.010505,-0.001872,0.002968,-0.001849,-0.006145,-0.00012

EDIT: If you're talking about what calculator traces of it appear on, I'll have to look through them again.

GenoPlot
06-16-2020, 10:24 PM
On various calculators I've seen small amounts of Iranian populations appear, so now exams have finished I thought I'd do some quick modelling on GenoPlot to see how certain populations fit out of interest. I don't know the region well enough to know how they might fit in with my known ancestry (Ashkenazi?), but I thought I'd share regardless. If I use Vahaduo and go FreeForm using specific samples, the Seyyed % jumps a bit. I haven't checked other Ashkenazi members to see if its the norm or not. If any of our members from the region have some thoughts, you're more than welcome to share. If its anything real, it certainly a very long way back.


Seeing that you had made your sample public, I took the liberty of running an optimaFit cycle against your base ancestry + all Iranian populations. In order to validate the presence of actual Iranian ancestry as opposed to traces of Levantine/Ashkenazi ancestry, I added a few other populations from the region. The resulting pool consisted of:

[ 'Irish', 'Croatian', 'English_Cornwall','Azeri', 'Iranian_Bandari', 'Iranian_Fars', 'Iranian_Jew', 'Iranian_Lor', 'Iranian_Mazandarani', 'Iranian_Seyyed', 'Iranian_Zoroastrian', 'Assyrian', 'Iraqi_Jew', 'Kurdish', 'Balochi', 'Samaritan', 'Lebanese_Christian','Syrian', 'Syrian_Jew' ]

and the results after evaluating the 3876 possible models?

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.189
Results:Irish71.2
English Cornwall14.4
Croatian9.6
Lebanese Christian4.8

BalkanKiwi
06-16-2020, 10:37 PM
Seeing that you had made your sample public, I took the liberty of running an optimaFit cycle against your base ancestry + all Iranian populations. In order to validate the presence of actual Iranian ancestry as opposed to traces of Levantine/Ashkenazi ancestry, I added a few other populations from the region. The resulting pool consisted of:

[ 'Irish', 'Croatian', 'English_Cornwall','Azeri', 'Iranian_Bandari', 'Iranian_Fars', 'Iranian_Jew', 'Iranian_Lor', 'Iranian_Mazandarani', 'Iranian_Seyyed', 'Iranian_Zoroastrian', 'Assyrian', 'Iraqi_Jew', 'Kurdish', 'Balochi', 'Samaritan', 'Lebanese_Christian','Syrian', 'Syrian_Jew' ]

and the results after evaluating the 3876 possible models?

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.189
Results:Irish71.2
English Cornwall14.4
Croatian9.6
Lebanese Christian4.8

Thanks for that. What a useful tool! So essentially the Iranian acts as a proxy of sorts for Levantine/Ashkenazi, when a better sample like Lebanese Christian isn't being used in the model? I assume this isn't to say I have actual ancestry from Lebanon specifically (that would be difficult to confirm), but more so its basically just representing the Levantine component of Ashkenazi?

GenoPlot
06-17-2020, 04:54 PM
Thanks for that. What a useful tool! So essentially the Iranian acts as a proxy of sorts for Levantine/Ashkenazi, when a better sample like Lebanese Christian isn't being used in the model? I assume this isn't to say I have actual ancestry from Lebanon specifically (that would be difficult to confirm), but more so its basically just representing the Levantine component of Ashkenazi?

The best way to look at it is that the mathematical probability for the origin of your West Asian ancestry is higher for a Levantine source when compared to an Iranian source.

The optimaFit process is designed to give users a good starting point for exploring their ancestry without the burden of manually creating non over-fitted models. In your case, if we take the automated model as the starting point and experiment further we can get even lower fits.

Turning the default penalty off:

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.0771
Results:Irish72
Croatian14
English Cornwall8
Lebanese Christian6

Substituting Ashkenazi for Lebanese:

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.084
Results:Irish73.5
Croatian10.5
Ashkenazi Russia10
English Cornwall6

The results in either case are close enough to be equivalent.

I would caution against reading too much into the results of any one individual sample. There is enough variation, even among siblings, that if the aim is to maximize accuracy, then results should be evaluated grouped with siblings and parents.

BalkanKiwi
06-17-2020, 11:34 PM
The best way to look at it is that the mathematical probability for the origin of your West Asian ancestry is higher for a Levantine source when compared to an Iranian source.

The optimaFit process is designed to give users a good starting point for exploring their ancestry without the burden of manually creating non over-fitted models. In your case, if we take the automated model as the starting point and experiment further we can get even lower fits.

Turning the default penalty off:

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.0771
Results:Irish72
Croatian14
English Cornwall8
Lebanese Christian6

Substituting Ashkenazi for Lebanese:

Sample:Balkan Kiwi ► BalkanKiwi
Fit:1.084
Results:Irish73.5
Croatian10.5
Ashkenazi Russia10
English Cornwall6

The results in either case are close enough to be equivalent.

I would caution against reading too much into the results of any one individual sample. There is enough variation, even among siblings, that if the aim is to maximize accuracy, then results should be evaluated grouped with siblings and parents.

I've been patiently waiting for the Eurogenes store to reopen in July, so I can get coordinates for my grandfather who is the source of the Ashkenazi. I'm hoping his data might provide some more clues. I can also get coordinates for my sister and mother as well depending on what his shows. It would be interesting to test this tool's ability to narrow done a sample using multiple family members.

DMXX
06-29-2020, 03:41 AM
The Iranian results look like a modelling artifact - Iranians and Ashkenazi Jews, broadly speaking, are in a similar zone ancestry-wise if we take a reductionist view of W. Eurasian ancestry (i.e. construing all ancestral pops as a mix of Villabruna-like HG's + BE-heavy "core" early W Asian).

Ashkenazi Jews model surprisingly well as a simple two-way mixture of Italian Tuscans and Iranians from Fars province:

[
{
"sample": "Ashkenazi Poland:Average",
"distance": 2.4955,
"Italian_Tuscany": 73,
"Iranian_Fars": 27
}
]

Clearly, this model is temporally false (and historically unlikely), but it emphasises the fairly intriguing outputs that nMonte occasionally produces.

I suspect that nMonte preferred having the quasi-Tuscan element of your Ashkenazi heritage as being more related to your other ancestries (e.g. Croatian), forcing the algorithm to find an ancestral source that best approximates the remainder (something W. Asian with a significantly larger proportion of Iran_N-derived ancestry).

I'd stumbled upon a similar situation myself with nMonte a couple years back - The addition of E. Slavs to my known ancestral pops consistently improved the fit and yielded 3-5% admix scores. A sign of cryptic E. Slavic admixture, or nMonte preferentially fitting them in to account for me being in the top centile of NW-N Iranian-descent folks when it comes to MLBA steppe admix? Segment sharing indicated the latter (I don't appear to have any unusually strong affinity for Ukrainians or Russians relative to other Iranians).

BalkanKiwi
06-29-2020, 04:42 AM
The Iranian results look like a modelling artifact - Iranians and Ashkenazi Jews, broadly speaking, are in a similar zone ancestry-wise if we take a reductionist view of W. Eurasian ancestry (i.e. construing all ancestral pops as a mix of Villabruna-like HG's + BE-heavy "core" early W Asian).

Ashkenazi Jews model surprisingly well as a simple two-way mixture of Italian Tuscans and Iranians from Fars province:

[
{
"sample": "Ashkenazi Poland:Average",
"distance": 2.4955,
"Italian_Tuscany": 73,
"Iranian_Fars": 27
}
]

Clearly, this model is temporally false (and historically unlikely), but it emphasises the fairly intriguing outputs that nMonte occasionally produces.

I suspect that nMonte preferred having the quasi-Tuscan element of your Ashkenazi heritage as being more related to your other ancestries (e.g. Croatian), forcing the algorithm to find an ancestral source that best approximates the remainder (something W. Asian with a significantly larger proportion of Iran_N-derived ancestry).

I'd stumbled upon a similar situation myself with nMonte a couple years back - The addition of E. Slavs to my known ancestral pops consistently improved the fit and yielded 3-5% admix scores. A sign of cryptic E. Slavic admixture, or nMonte preferentially fitting them in to account for me being in the top centile of NW-N Iranian-descent folks when it comes to MLBA steppe admix? Segment sharing indicated the latter (I don't appear to have any unusually strong affinity for Ukrainians or Russians relative to other Iranians).

This makes sense (in your case as well regarding E. Slavs). Ignoring the distance on that model, its surprising how much is assigned to Fars, however if its a broad representation of West Asian ancestry, it should probably be expected. If anything, its a good example of what nMonte will do in order to create somewhat of a "realistic" model based on the entered data. As we know, the statistical test doesn't know if the output is right or wrong.

iloko
08-16-2020, 03:03 AM
Is NMonte Runner based on the FST method, or the LeastSquares??

IronHorse
05-24-2021, 09:39 PM
this is mine, Iranian improves the fit a lot, I wonder why


Sample:Prometheus
Fit:0.9473
Results:BedouinB38
Yemenite Al Jawf25.5
Iranian18
Yemenite Mahra15.5
Somali3


the second best is with Iraqi


Sample:Prometheus
Fit:1.1377
Results:BedouinB33.5
Iraqi30.5
Yemenite Al Jawf19.5
Yemenite Mahra15
Somali1.5



edit: after playing a little bit more, Tajik is better


Sample:Prometheus
Fit:0.8982
Results:BedouinB49.5
Yemenite Al Jawf28.5
Tajik12
Yemenite Mahra9
Somali1