Single thread for G25 modelling requests

[JerryS.]

Ive been saying that you need to make targeted, well educated reference pops to run, and not just throw random modern pops into a spreadsheet and think its accurate. An EEf signal can be incorrectly modeled as many different things if you are not careful.




Jerry, at this point I feel like I'm trolling you, lol. I've made multiple comments about this but you never reply. You have multiple commercial tests showing a 95% +/- NW Euro make up. Before the latest v5.9, 23 and me said you were what, 5% Italian? But the more telling thing was your #1 ancestor location was Veneto, or something like that. So what are you expecting the G25 results to show? My aunt, who scores 55-65% Southern Italian on multiple commercial tests always shows as Tuscan, or Lombard in G25. My daughter, who scores a whopping (/s) combined 4% or 5% Italian on 23 and me has Calabria as her #1. She's an 1/8th Italian, same as you. I think you should explore other area's of the peninsula, or work on solidifying the paper trail before you take the G25 results as facts.

I've posted elsewhere (the tellmegen thread as the latest) that the G25 data and the other free ethnicity sites should be taken for fun when compared to commercial test where actual spit samples were used, that seems to be your view in a nutshell right?

[digital_noise]

I've posted elsewhere (the tellmegen thread as the latest) that the G25 data and the other free ethnicity sites should be taken for fun when compared to commercial test where actual spit samples were used, that seems to be your view in a nutshell right?

Yes and no. These tests, as well as the older gedmatch calcs often use the same sample pool, so the results will often align somewhat. But If you have multiple commercial tests, the results will be different. That's just a given. But the commercial tests will paint an overall bigger picture of what is going on. So if TellemGen, or Ethnogenes comes out with some result from left field, I would take that with a grain of salt, especially if they dont divulge their method. same can be said with G25. If some random sample pops are thrown together, then the end result may or may not align with what is actually the truth. Another thing, G25 can seriously mislead you if you are not careful or knowledgeable. If someone has SE European ancestry, that can easily be manipulated to other ethnicities and still get a decent enough fit, so if someone has no idea of their background, its really easy to be misled. Thats just misuse of the tool however, not a fault of it.

[JerryS.]

Yes and no. These tests, as well as the older gedmatch calcs often use the same sample pool, so the results will often align somewhat. But If you have multiple commercial tests, the results will be different. That's just a given. But the commercial tests will paint an overall bigger picture of what is going on. So if TellemGen, or Ethnogenes comes out with some result from left field, I would take that with a grain of salt, especially if they dont divulge their method. same can be said with G25. If some random sample pops are thrown together, then the end result may or may not align with what is actually the truth. Another thing, G25 can seriously mislead you if you are not careful or knowledgeable. If someone has SE European ancestry, that can easily be manipulated to other ethnicities and still get a decent enough fit, so if someone has no idea of their background, its really easy to be misled. Thats just misuse of the tool however, not a fault of it.

so, just selecting ALL of the individual samples from specific countries, and all the samples from bordering regions of those countries, as well as historically contributing countries, leaving out all others like like Romanian, Iceland, Chinese, Sammi, Bantu... is not a good practice? *** for G25 use on Vahaduo I mean, like what I did in the earlier post.

[digital_noise]

so, just selecting ALL of the individual samples from specific countries, and all the samples from bordering regions of those countries, as well as historically contributing countries, leaving out all others like like Romanian, Iceland, Chinese, Sammi, Bantu... is not a good practice? *** for G25 use on Vahaduo I mean, like what I did in the earlier post.

You’d likely get over fitted results. I don’t bother removing extreme outliers like Chinese etc so I cannot comment. Just run your results against the whole sheet, While likely overfitted, it shouldn’t look too foreign to you.

[JerryS.]

You’d likely get over fitted results. I don’t bother removing extreme outliers like Chinese etc so I cannot comment. Just run your results against the whole sheet, While likely overfitted, it shouldn’t look too foreign to you.

I did that on the previous page with the reduced to 4 pops so the fractional populations can be consolidated instead of 21 additional populations of 0.4%....

[J Man]

I know that this is revisiting an old topic but since using ancient references in G25 is still good for admixed European/West Eurasian individuals such as myself I thought that I would ask for some advice. I am curious about how much Steppe, EEF, WHG, Post Neolithic Iran, Siberian etc. ancestry I have.

I would like to figure out a good model/test for my G25 to use on vahaduo. My ancestral breakdown is like this. I am 25% Italian (Calabria region), 25% Finnish and 50% British Isles by known ancestry. Can anyone here please suggest which ancient reference populations I should use to figure out how much Steppe, EEF, WHG, Post-Neolithic Iran, Siberian and other ancestral components I have?

[Leper]

Don't treat G25 like ADMIXTURE and don't use the whole speadsheet when running a model. G25 is meant to be used in supervised way. You'll get very accurate results, you just need to know what you are doing first.

[JerryS.]

Don't treat G25 like ADMIXTURE and don't use the whole speadsheet when running a model. G25 is meant to be used in supervised way. You'll get very accurate results, you just need to know what you are doing first.

I'd like to try that supervised way one day.

[digital_noise]

All that means is using well chosen sample populations in a calculator. Samples that don’t over fit, overlap etc...

[J Man]

My results with the Standard Global25 Calculator by Davidski at Vahaduo.

Target: J Man_scaled
Distance: 3.0532% / 0.03053157
48.6 Yamnaya_RUS_Samara
39.8 TUR_Barcin_N
10.4 WHG
1.0 Han
0.2 Nganassan