Saturday, September 10, 2022

The difference between scaled and unscaled in Vahaduo/G25.

 The difference between scaled and unscaled in Vahaduo/G25.

When using G25 coordinates one important thing to keep in mind is that the G25 is not directly working on any alleles. G25 zips any ancient DNA into 25 numbers. Which are coordinates in a virtual 25 dimension PCA.
The application that makes them creates unscaled coordinates first. Unscaled "humanity" is much closer to each other than the scaled one. As You can see in unscaled mode Turks and Azerbaijani are quite close to Armenians than in the scaled. The closeness of Turkish pops makes sense from historic point of view. In this sense the unscaled G25 data looks more close to Fst.
So why they are not close in the scaled? The reason of this is that scaling makes divergent pops like East Asian , African, Amerindian, Papuan deliberately more distant from West Eurasians. As consequence an East Turkish pop who have less than 10% East Asian is pulled away from Armenians while the local Near Easterners who probably have less shared ancestry with Armenians becomes apparently closer.
So when You use the scaled coordinates keep in mind this feature. This is even more important for ancient DNA. Many aDNA suffer from damage. They have noise levels of various anachronistic ancestry from Africa, America, East Asia which will affect the distances also.
In sum unscaled behavior is closer to Fst while the scaled behavior is more close to Admixture based oracles.
PS I used Erzurum because it's close to average.

All reactions:

No comments:

Post a Comment