Global25 nMonte and PCA's

[Gründig]

This is with my converted V3 data, which is apparently comparable to FTDNA raw data (one of the best DNA tests for raw data).

Heres are two pretty solid fits between .5% and .9% using nMonte with R:

Unscaled individuals and i took out all the known southern Lombard samples.

"distance%=0.7547"

AS_G25

Italy_Medieval_Collegno,51.2 (Lombards)
Hungary_Medieval_Szolad,37.6 (Lombards)
Germany_Medieval,8.6 (Germanic Bavarian)
Sweden_Viking_Age,2.6


Heres a pretty simple run with modern unscaled individual populations:

"distance%=0.6157"

AS_G25

German,67.8
Dutch,15
French,13.2
Austrian,4

[Gründig]

Celtic vs Germanic:

"John_Smith" black dot

http://i.imgur.com/I5Ot31Q.jpg

[Gründig]

I figured more people would have done the global25.

[Teutonski]

Übermensch DNA :thumb001:

[Gründig]

Übermensch DNA :thumb001:

The tests are pretty cool. You can compare yourself against ancient populations, like I did at the top with the lombard Germanic tribes.

The title "Hungary_Medieval_Szolad" and "Italy_Medieval_Collegno" can be misleading. Those were just the locations these samples were found.

[Teutonski]

The tests are pretty cool. You can compare yourself against ancient populations, like I did at the top with the lombard Germanic tribes.

The title "Hungary_Medieval_Szolad" and "Italy_Medieval_Collegno" can be misleading. Those were just the locations these samples were found.

How you did the test?

[Gründig]

How you did the test?

You send your raw data to David at eurogenes and he sends you back coordinates.

You can than do it yourself with these coordinates through a program called R or you can do it through this guy poi's online nMonte calculator.

[Teutonski]

You send your raw data to David at eurogenes and he sends you back coordinates.

You can than do it yourself with these coordinates through a program called R or you can do it through this guy poi's online nMonte calculator.

complicated?

[Gründig]

complicated?

With using program R, it can be at first.

With poi's calculator, it's much easier. You're just more limited but it's still great.

I've learned to do both, so if you ever decide to get the coordinates ($12), I could help you out.

[Gründig]

I just found out using scaled coordinates for the models is best. The distance to aim for is between 1% and 2%.

Heres an ancient scaled run with all the southern Lombards removed:

"distance%=1.4404"

AS_G25_scaled

Hungary_Medieval_Szolad,45.8 (Lombards)
Italy_Medieval_Collegno,34.6 (Lombards)
Germany_Medieval,18.2 (Bavarian Germanic)
Sweden_Viking_Age,1.4

[Chaos One]

I did it with K36 Ancient nMonte, got nice results. The archives that I found had like 10 different populations to mix, so you could get from HG to Chalcolithic or Bronze Age to High Medieval Age, etc.

[Gründig]

I did it with K36 Ancient nMonte, got nice results. The archives that I found had like 10 different populations to mix, so you could get from HG to Chalcolithic or Bronze Age to High Medieval Age, etc.

What are these archives?

Global25 has a decent amount of samples.

[Grace O'Malley]

This is with my converted V3 data, which is apparently comparable to FTDNA raw data (one of the best DNA tests for raw data).

Heres are two pretty solid fits between .5% and .9% using nMonte with R:

Unscaled individuals and i took out all the known southern Lombard samples.

"distance%=0.7547"

AS_G25

Italy_Medieval_Collegno,51.2 (Lombards)
Hungary_Medieval_Szolad,37.6 (Lombards)
Germany_Medieval,8.6 (Germanic Bavarian)
Sweden_Viking_Age,2.6


Heres a pretty simple run with modern unscaled individual populations:

"distance%=0.6157"

AS_G25

German,67.8
Dutch,15
French,13.2
Austrian,4

Can you model me using modern populations and then ancients, if possible? If not let me know. I have access to poi's tool but you have to select the populations on that. Thanks in advance.

,PC1,PC2,PC3,PC4,PC5,PC6,PC7,PC8,PC9,PC10,PC11,PC1 2,PC13,PC14,PC15,PC16,PC17,PC18,PC19,PC20,PC21,PC2 2,PC23,PC24,PC25
Grace_scaled,0.138864,0.136081,0.064488,0.052003,0 .036314,0.017291,0.001645,0.006231,0.004909,-0.00328,-0.009094,0.004646,-0.011298,-0.009771,0.029994,0.005701,-0.007041,-0.004434,0.000377,0.007879,0.007112,0.007296,-0.005793,0.010845,-0.000479

,PC1,PC2,PC3,PC4,PC5,PC6,PC7,PC8,PC9,PC10,PC11,PC1 2,PC13,PC14,PC15,PC16,PC17,PC18,PC19,PC20,PC21,PC2 2,PC23,PC24,PC25
Grace_unscaled,0.0122,0.0134,0.0171,0.0161,0.0118, 0.0062,0.0007,0.0027,0.0024,-0.0018,-0.0056,0.0031,-0.0076,-0.0071,0.0221,0.0043,-0.0054,-0.0035,0.0003,0.0063,0.0057,0.0059,-0.0047,0.009,-0.0004

[Gründig]

Can you model me using modern populations and then ancients, if possible? If not let me know. I have access to poi's tool but you have to select the populations on that. Thanks in advance.

,PC1,PC2,PC3,PC4,PC5,PC6,PC7,PC8,PC9,PC10,PC11,PC1 2,PC13,PC14,PC15,PC16,PC17,PC18,PC19,PC20,PC21,PC2 2,PC23,PC24,PC25
Grace_scaled,0.138864,0.136081,0.064488,0.052003,0 .036314,0.017291,0.001645,0.006231,0.004909,-0.00328,-0.009094,0.004646,-0.011298,-0.009771,0.029994,0.005701,-0.007041,-0.004434,0.000377,0.007879,0.007112,0.007296,-0.005793,0.010845,-0.000479

,PC1,PC2,PC3,PC4,PC5,PC6,PC7,PC8,PC9,PC10,PC11,PC1 2,PC13,PC14,PC15,PC16,PC17,PC18,PC19,PC20,PC21,PC2 2,PC23,PC24,PC25
Grace_unscaled,0.0122,0.0134,0.0171,0.0161,0.0118, 0.0062,0.0007,0.0027,0.0024,-0.0018,-0.0056,0.0031,-0.0076,-0.0071,0.0221,0.0043,-0.0054,-0.0035,0.0003,0.0063,0.0057,0.0059,-0.0047,0.009,-0.0004

I actually learned that scaled is the way to go, so I was going to run some more on myself after work. I can do some for you as well.

I know you're Irish but are you anything else? Just so I have a direction to go in.

[Grace O'Malley]

I actually learned that scaled is the way to go, so I was going to run some more on myself after work. I can do some for you as well.

I know you're Irish but are you anything else? Just so I have a direction to go in.

Thank you. All my ancestry is Irish. Do you have to actually pick populations to model yourself on or does the program pick what matches you best?

[Gründig]

Thank you. All my ancestry is Irish. Do you have to actually pick populations to model yourself on or does the program pick what matches you best?

You have to pick all the populations you want to compare yourself to manually.

[Grace O'Malley]

You have to pick all the populations you want to compare yourself to manually.

So it is similar to using poi's tool. My closest population is actually Icelandic on this oddly enough so I usually get a chunk of Icelandic. I've had people do modelling for me previously. All my ancestry is Irish but I usually get some Scandinavian which could be deep ancestry because I have no known recent ancestors from there.

[Gründig]

So it is similar to using poi's tool. My closest population is actually Icelandic on this oddly enough so I usually get a chunk of Icelandic. I've had people do modelling for me previously. All my ancestry is Irish but I usually get some Scandinavian which could be deep ancestry because I have no known recent ancestors from there.

It is, you can just run more populations without paying. The distances also seem to vary a little between the two programs but nothing crazy.

Yea I can get some closer distances with some nonsense combinations. However, sometimes certain populations may have samples that fall within another population. For example, it seems I'm closest to a french sample but when I run the test, German always comes out on top. In other words, that French sample clusters with the Germans.

[Token]

You are overfitting.

[Grace O'Malley]

This is what I get for Irish, English and Icelandic

Fit 0.8221
English 5.83
Icelandic 31.67
Irish 62.5

But I get a closer distance by substituting Dutch for English

Fit 0.7667
Dutch 17.5
Icelandic 29.17
Irish 53.33

So obviously the closer the fit the better but I doubt I really have Icelandic ancestry or Dutch but I don't match England very well.

Your above population match is a really low distance. Is that your actual known ancestry because it looks a really good model for you?

[Grace O'Malley]

You are overfitting.

Is that using populations that are too similar?

[Token]

Is that using populations that are too similar?

Basically.

[Gründig]

You are overfitting.

For which? And what populations am I supposed to use then? The moderns are my actual ancestry and the ancients align with that pretty well.

[Grace O'Malley]

Basically.

Could you give an example of a good model to use? People have to go with their known ancestry as well.

[Grace O'Malley]

If I go with the populations from the Irish DNA Atlas re French and Norwegian making up the majority of the Irish genepool this is what I get.

Fit 1.6527
French_Average 31.67
Norwegian 68.33

[Gründig]

Could you give an example of a good model to use? People have to go with their known ancestry as well.

Yea I've seen different definitions of over fitting so I'm not really sure what it means.

I learned the key is to use realistic samples of your known ancestry and try staying within the 4-5 population amount. I'm not really sure what I could do differently. I meet those constraints.

I could get a decent amount closer but it would become unrealistic.

[Chaos One]

What are these archives?

Global25 has a decent amount of samples.

https://anthrogenica.com/showthread.php?11663-K36-based-oracle-with-ancients/page13

[Chaos One]

For example, my Bronze Age to Early Medieval results:

BA_Hungary_RISE254 22.45
BA_I9041_Mycenaean 16.35
Egyptian_mumy_I_JK2888_ 15.10
Amerindian_Kennewick 12.75
IA_EastKazachstan_Is2 8.95
IA_Wielbark_Kow_26_PL 6.30
MBA_ATP9_Iberia 5.80
LBA_Armenia_RISE396 4.30
BA_Hungary_RISE371 3.20
BA_Hungary_RISE374 3.00
HG_Ethiopia_Mota1 1.80

[Imperator Biff]

I just found out using scaled coordinates for the models is best. The distance to aim for is between 1% and 2%.

The optimum target distance is actually more so around 4%, anything lower than that means the model is overfitted.

[Gründig]

The optimum target distance is actually more so around 4%, anything lower than that means the model is overfitted.

This is false. I've spoken to David about it before. For scaled coordinates, a solid fit is between 1% to 2% with 4 to 5 populations being used. He did say that distances up in the 4% area can sometimes be acceptable as well.

If what you were saying was true, I would be over fitting drastically using the Germany populations alone....

[Raizen]

This is false. I've spoken to David about it before. For scaled coordinates, a solid fit is between 1% to 2% with 4 to 5 populations being used. He did say that distances up in the 4% area can sometimes be acceptable as well.

If what you were saying was true, I would be over fitting drastically using the Germany populations alone....

does a models like that looks solid? i think it fits pretty well my background

"distance%=2.268"

Portuguese_Madeira,78.8
Yoruban,12
Ashkenazi,6
Karitiana,3.2

[Token]

For which? And what populations am I supposed to use then? The moderns are my actual ancestry and the ancients align with that pretty well.

You are using outgroups with too much overlap, specially on your first model. The Collegno and Szólad outgroups, for example, are basically identical. You shoud aim 1-2%, anything lower than that and you are already overfitting your model.

By the way, Global25 is not the best for modelling recent ancestry, their spreadsheet is severely lacking on this matter.

[Grace O'Malley]

You are using outgroups with too much overlap, specially on your first model. The Collegno and Szólad outgroups, for example, are basically identical. You shoud aim 1-2%, anything lower than that and you are already overfitting your model.

By the way, Global25 is not the best for modelling recent ancestry, their spreadsheet is severely lacking on this matter.

What's the best for modelling recent ancestry?

[Token]

What's the best for modelling recent ancestry?

K13 or K36 spreadsheet. You need to do everything manually but it compensates you with much more flexibility. G25 is completely focused on ancient samples.

[Gründig]

K13 or K36 spreadsheet. You need to do everything manually but it compensates you with much more flexibility. G25 is completely focused on ancient samples.

Didn't David create eurogenes as well? The results are that different? My gedmatch K13 actually aligns pretty well with the nMonte global25 runs I did.

Either way, how do you go about running the k13 or k36 spreadsheets through nMonte?


You are using outgroups with too much overlap, specially on your first model. The Collegno and Szólad outgroups, for example, are basically identical. You shoud aim 1-2%, anything lower than that and you are already overfitting your model.

By the way, Global25 is not the best for modelling recent ancestry, their spreadsheet is severely lacking on this matter.

Those first runs at the top of this post were using unscaled samples (I recently learned that I should use scaled) and using one sample along would already put me at like below a 1.5 distance. So I asked around and was told with unscaled to aim for .5% to .9%. So my runs were within the constraints. However, using the scaled you aim for 1% to 2% as you said.

[Grace O'Malley]

This is most likely the most accurate as far as what could contribute to my ancestry but not the lowest number. I'm not sure what else I can accurately add.

Fit 0.8652
Irish 70.83
Norwegian 17.5
Scottish 11.67

If I add French to this I get 0.

Fit 0.8647
French 0
Irish 70.83
Norwegian 17.5
Scottish 11.67

Not really sure of what else I can add realistically.

If I remove Irish it mostly goes to Scottish.

Fit 1.2681
French 5
Norwegian 24.17
Scottish 70.83

[Gründig]

This is most likely the most accurate as far as what could contribute to my ancestry but not the lowest number. I'm not sure what else I can accurately add.

Fit 0.8652
Irish 70.83
Norwegian 17.5
Scottish 11.67

If I add French to this I get 0.

Fit 0.8647
French 0
Irish 70.83
Norwegian 17.5
Scottish 11.67

Not really sure of what else I can add realistically.

If I remove Irish it mostly goes to Scottish.

Fit 1.2681
French 5
Norwegian 24.17
Scottish 70.83

This is all on the online tool? I specifically asked David what to aim for using this tool and he said 4-5 populations and between 1% to 2%.

So that looks good. The fact you're below 1% in this case would be most likely considered "over fitting".

Sorry I haven't run your results yet, things got really busy yesterday.

[Gründig]

does a models like that looks solid? i think it fits pretty well my background

"distance%=2.268"

Portuguese_Madeira,78.8
Yoruban,12
Ashkenazi,6
Karitiana,3.2

Was this on poi's tool? Or nMonte with R? scaled or unscaled ?

[Grace O'Malley]

This is what Aren kindly did for me previously. It most likely is a better model than I've done for myself.

"distance%=1.7153"

Grace_scaled

Irish,55.2
Welsh,24.2
Swedish,20.6

"distance%=1.6364"

Grace_scaled

Icelandic,44.4
Irish,29.6
Welsh,26

[Token]

Was this on poi's tool? Or nMonte with R? scaled or unscaled ?
I ran this model for him with nMonte + K13. The advantage was that i could choose between several Portuguese regions, something that is not possible with G25. In his case, using Central Portuguese or Madeiran samples improved the fit significantly. I couldn't get past 2.25% fst, probably because he is too mixed.

[Gründig]

I ran this model for him with nMonte + K13. The advantage was that i could choose between several Portuguese regions, something that is not possible with G25. In his case, using Central Portuguese or Madeiran samples improved the fit significantly. I couldn't get past 2.25% fst, probably because he is too mixed.

How are you doing this? I would like to try.

I know with G25 you use the coordinates in a target file and pop data in a different file but I'm not sure how to set it up without using these.

[Gründig]

Here's one of David's PCA's. I added the colors.

Black star = Me
Yellow = German
Green = Dutch
Dark blue = East French
Unfilled squares = Hungary_Medieval_Szolad (Lombard)
Filled squares = Italy_Medieval_Collegno (Lombard)

My dot is overlapped by the Germans, Dutch and just about East French. I'm also surrounded by Lombard samples.


http://i.imgur.com/kHM89zH.jpg

[Gründig]

I'm the black + sign:

http://i.imgur.com/Z1s9yxR.jpg

[Gründig]

Bump

[Gründig]

I had this K13 nMonte run for me. It seems they used all the populations, hence why there is so many small numbers. I cut off anything under 1%.

I would like to have a run done with only 4-5 populations.


West_German South_Dutch Austrian East_German North_German French
0.3901679 0.5222337 0.6044510 0.8247824 0.9332004 0.9721312
Southeast_English Hungarian Danish North_Dutch
1.1061370 1.1110090 1.1745510 1.1901949

"distance%=0.1088" (1.088)

West_German,39.4
South_Dutch,9.2
Austrian,4.2
North_German,3.8
East_German,3.2
Norwegian,3.2
Southwest_English,2.2
North_Dutch,2
Hungarian,1.8
Irish,1.8
Spanish_Cataluna,1.8
South_Polish,1.6
West_Scottish,1.6
Romanian,1.2
Belorussian,1
French,1

[Gründig]

Heres another K13 nMonte that the same person ran for me. Looks like all the populations were used again, so I cut out anything lower than 1%.

Using pen=0

"distance%=0.0448" (0.448)

Austrian,23
North_Dutch,16.2
Norwegian,13.8
South_Dutch,12.2
South_Polish,6.2
Serbian,6
North_Italian,5.6
Spanish_Cataluna,4.2
Danish,2.8
Tuscan,2.2
French_Basque,2
Southwest_English,1.8


I'm not positive how pen=0 works. The K13 and K13 pen=0 have their similarities but many differences as well. A couple strange things popped up in the pen=0. Interesting nonetheless.