August 25, 2012

ADMIXTURE 1.22 correcting Fst bias

In many of my experiments I have used ADMIXTURE versions prior to 1.22. According to the authors' website, version 1.22 (3/10/2012):
Fst estimates were upward biased; have now switched to the method of Reynolds et al. (1983).
This probably means that many of the Fst divergences reported here and in the Dodecad blog must be reduced. This is not really a big problem, since, biased or not, the reported numbers show the relative similarity of difference components. But, I decided to investigate, so I re-ran the ADMIXTURE analysis that created the K7b calculator.


The correlation with the old (ver. 1.21) Fst values is very strong (+0.9993209) and the new values can be estimated from the old using the following regression:

New = 0.782324*Old + 0.009335

Of course it would be a good idea to re-run this type of analysis separately whenever the absolute values are important. For example, in a previous experiment, I suggested that Fst's between the K12a components were so low, that these components should not be interpreted as having diverged in very old (say, Upper Paleolithic) times, but rather in a more recent (post-glacial, and probably mostly Neolithic) time frame. Correction for this upward bias would probably strengthen that hypothesis which was one way of arguing in favor of the womb of nations theory.

2 comments:

Anonymous said...

West Eurasian, East Eurasian and African groups seem to have shifted closer together relative to populations within those groups (based on the original K7 Fst's).

Nathan Paul said...

So is it South Asian Womb of nations as it is Equidistant to all the regions except Africa?