Compare "real" G25 coordinates to Simulated-G25 results with K36

Unsupervised (?) Admixture analysis overweights Finnish proportion, which is typical for Markov chain or similar calculations. The error is not caused by the simulation.

Target: sample_simulated_g25_scaled
Distance: 1.1339% / 0.01133885 | ADC: 1x RC
32.6 Finnish_Central
24.2 Estonian
22.0 Ingrian
21.2 Finnish_Southeast

Target: MauriJussi1m_scaled
Distance: 1.3766% / 0.01376590 | ADC: 1x RC
55.2 Ingrian
25.2 Swedish
19.6 Finnish_Southwest

Target: sample_simulated_g25_scaled
Distance: 1.0519% / 0.01051894 | ADC: 0.5x RC
31.8 Finnish_Southeast
25.6 Ingrian
22.2 Finnish_Southwest
18.0 Estonian
1.2 Finnish_Central
1.2 Swedish

Target: MauriJussi1m_scaled
Distance: 1.2509% / 0.01250924 | ADC: 0.5x RC
36.8 Finnish_Central
22.4 Swedish
20.8 Russian_Tver
10.8 Ingrian
9.2 Finnish_Southwest

Target: sample_simulated_g25_scaled
Distance: 1.0189% / 0.01018895 | ADC: 0.25x RC
33.4 Finnish_Southeast
23.2 Finnish_Southwest
20.4 Ingrian
13.2 Latvian
7.2 Finnish_Central
2.6 Swedish

Target: MauriJussi1m_scaled
Distance: 0.8252% / 0.00825213 | ADC: 0.25x RC
41.8 Finnish_Central
21.6 Lithuanian_VZ
21.4 Shetlandic
7.6 Ingrian
5.8 Russian_Yaroslavl
1.4 Murut
0.4 Maori

My regular G25 coordinates were always very bad giving me huge distances to my own ethnicity (and I'm not the only one with this issue). This is a much needed tool that really improved my coordinates on G25. Can't believe I paid for G25 three times for each raw data file.

Distance to: Mingle_raw_data
0.03150615 Pashtun_Uthmankhel
0.03411736 Kalash
0.03544066 Pashtun_Tarkalani
0.03575387 Pashtun_Yusufzai
0.03970679 Pashtun_North_Afghanistan
0.04094886 Pashtun_Kurram_o
0.04267163 Kamboj
0.04577366 Balochi_B
0.04729382 Kho_Singanali
0.05003843 Arora
0.05207452 Ror
0.05405719 Khatri
0.05414879 Jat_Uttar_Pradesh
0.05465078 Jat_Haryana
0.05492211 Brahui
0.05558596 Jat_Punjab_Sikh
0.05571883 Jat_Punjab_Muslim
0.05674197 Tajik_Ishkashim
0.05717209 Arain
0.05847219 Kohistani
0.06005915 Tajik_Wakhi
0.06114986 Tajik_Badakhshan
0.06167490 Punjabi_Sikh_India
0.06208694 Balochi_A
0.06227557 Sindhi



Distance to: Mingle_K36_simulated
0.01711584 Pashtun_Uthmankhel
0.01801670 Pashtun_Tarkalani
0.02197727 Pashtun_Yusufzai
0.02646254 Kalash
0.03005094 Pashtun_North_Afghanistan
0.03304623 Pashtun_Kurram_o
0.03510766 Kamboj
0.03918735 Balochi_B
0.03987141 Kho_Singanali
0.04428746 Arora
0.04604669 Brahui
0.04630223 Khatri
0.04707374 Ror
0.04863873 Jat_Uttar_Pradesh
0.04959366 Jat_Haryana
0.04990579 Tajik_Ishkashim
0.05056736 Jat_Punjab_Muslim
0.05096571 Jat_Punjab_Sikh
0.05291895 Kohistani
0.05297213 Arain
0.05362546 Balochi_A
0.05458751 Tajik_Wakhi
0.05574226 Sindhi
0.05711040 Tajik_Badakhshan
0.05724876 Punjabi_Sikh_India

Quote:

---

Originally Posted by Mingle

My regular G25 coordinates were always very bad giving me huge distances to my own ethnicity (and I'm not the only one with this issue). This is a much needed tool that really improved my coordinates on G25. Can't believe I paid for G25 three times for each raw data file.

Distance to: Mingle_raw_data
0.03150615 Pashtun_Uthmankhel
0.03411736 Kalash
0.03544066 Pashtun_Tarkalani
0.03575387 Pashtun_Yusufzai
0.03970679 Pashtun_North_Afghanistan
0.04094886 Pashtun_Kurram_o
0.04267163 Kamboj
0.04577366 Balochi_B
0.04729382 Kho_Singanali
0.05003843 Arora
0.05207452 Ror
0.05405719 Khatri
0.05414879 Jat_Uttar_Pradesh
0.05465078 Jat_Haryana
0.05492211 Brahui
0.05558596 Jat_Punjab_Sikh
0.05571883 Jat_Punjab_Muslim
0.05674197 Tajik_Ishkashim
0.05717209 Arain
0.05847219 Kohistani
0.06005915 Tajik_Wakhi
0.06114986 Tajik_Badakhshan
0.06167490 Punjabi_Sikh_India
0.06208694 Balochi_A
0.06227557 Sindhi



Distance to: Mingle_K36_simulated
0.01711584 Pashtun_Uthmankhel
0.01801670 Pashtun_Tarkalani
0.02197727 Pashtun_Yusufzai
0.02646254 Kalash
0.03005094 Pashtun_North_Afghanistan
0.03304623 Pashtun_Kurram_o
0.03510766 Kamboj
0.03918735 Balochi_B
0.03987141 Kho_Singanali
0.04428746 Arora
0.04604669 Brahui
0.04630223 Khatri
0.04707374 Ror
0.04863873 Jat_Uttar_Pradesh
0.04959366 Jat_Haryana
0.04990579 Tajik_Ishkashim
0.05056736 Jat_Punjab_Muslim
0.05096571 Jat_Punjab_Sikh
0.05291895 Kohistani
0.05297213 Arain
0.05362546 Balochi_A
0.05458751 Tajik_Wakhi
0.05574226 Sindhi
0.05711040 Tajik_Badakhshan
0.05724876 Punjabi_Sikh_India

---

Simulated coords are much less accurate.

You get high distances because that's the reality and you likely have some drift. Just because you have low distances with sim coords doesn't mean they are better.

They aren't. They are worse and far less accurate.

Entire G25 setup is made with real coords including all the source data. Which is why simulated coords will never come close. And low distances they give to everyone are pseudo and incorrect.

Quote:

---

Originally Posted by Feiichy

Simulated coords are much less accurate.

You get high distances because that's the reality and you likely have some drift. Just because you have low distances with sim coords doesn't mean they are better.

They aren't. They are worse and far less accurate.

Entire G25 setup is made with real coords including all the source data. Which is why simulated coords will never come close. And low distances they give to everyone are pseudo and incorrect.

---

It doesn't make sense for me to score closer to Kalashas than to most samples from my own ethnicity. Kalashas have their own separate profile and are more isolated & drifted from me. On every GEDmatch calc, I'm closer to other Pashtuns than to Kalashas for a reason.

And it doesn't make sense for me to be get such a high distance to my own people when you have Northern European users scoring much closer to Bell Beaker samples. Am I really more drifted from modern samples than N. Euro users are from ancient samples?

A distance above 0.3 for your closest population is just too much. I think G25 could be better than K36 simulated coordinates for most people, but it messes up for a few people on an individual level. G25 itself isn't perfect and is a PCA based model, that's why people are trying to use qpAdm more.

Quote:

---

Originally Posted by Feiichy

Simulated coords are much less accurate.

You get high distances because that's the reality and you likely have some drift. Just because you have low distances with sim coords doesn't mean they are better.

They aren't. They are worse and far less accurate.

Entire G25 setup is made with real coords including all the source data. Which is why simulated coords will never come close. And low distances they give to everyone are pseudo and incorrect.

---

for me they seem to be even better. I get better distances all around.