torstai 20. helmikuuta 2014

Looking closer: The Finnish cluster in Europe

New PCA's

A new PCA analysis including only Europeans.   The first plot including dimensions 1 and 2 shows for Finns East European and Scandinavian affinities, more precisely for the Eastern Finns the East European affinity closer eastern Vologda samples and the Western Finns match closer western Vologda samples and Belarussians.    Scandinavian affinity on second dimension looks quite constant for all Finns.   This result could mean that the common affinity between Finns and Scandinavians is older than the present genetic shape of these populations (Finns and Scandinavians) and the difference in the East European affinity among Finns idicates also different origins for Eastern and Western Finns in Eastern Europe. This explanation is supported also by the position of Scandinavian and some CEU samples who seem to diverge from Central Europeans toward Finns (dimension 2). The second plot, showing dimensions 1 and 3, confirms the East European affinity of Finns.  Scandinavians keep their position near Central Europeans. 

Full size image available here

Full size image available here

Retesting Finestructure

I have a new interesting experiment with Finestructure.   I played with it a couple years ago testing it several days and found some problems.  It seemed to exaggerate young and small isolated populations by increasing their similarities with other populations in straight relation to the shared haplotypes within the isolation.  This effect is a bit difficult to describe exactly by words although testing revealed it certainly. After seeing this problem I decided to give up with Finestructure.  Maybe the problem was partially in Chromopainter and it’s data selections.   Now, in my new experiment I used IBS data as an input to the Finestructure.  The authors of Finestructure claim that IBS data is much worse than the data generated by Chromopainter.  My results show that IBS data gives fair good results, better than in my previous tests with Chromopainter data which was pretty much distorted by the problem I described above.   Pretty much testing ought to be done before this can be proved stone cold, and even then many readers will not believe me now, so just now it is enough for me to know this.   I think that the effect is basically the same I have seen so many times in population genetic tests;   many analyses “infer” that most homogeneous populations, owning highest shared genome similarity, are root populations.  This means that those analyses often select typical genetic features from most homogeneous subpopulations.  But this is a big mistake; in many cases the most homogeneous subpopulation is younger than the root population, the one they once diverged from.     In other words:  the portrait doesn’t come better by increasing contrast, because we likely will lose nuances and sometimes also profile resembles the neighbor.    
So I spent a little time more with testing and I think that I found a way to avoid the false gene pumping between populations described above.   Definitely, this test done with IBS-data looks very clean.   Some problems can be seen in sorting out Mediterranean populations, but in Northern and Eastern Europe the grouping result is amazing.   The Finns cluster with Northern Russians, which is of course right and follows the known history.  Northwest Russia was the homeland for many Finnic speaking groups before the russification took place.    Tree model can’t figure simultaneously the Finns belonging also to the Fennoscandinavian group.    

The graphic shows also affinities between individuals.

Full size image available here

Ei kommentteja:

Lähetä kommentti