Tuesday, 22 March 2011

ADMIXTURE doesn't know better

I've been running ADMIXTURE recently, seeking evidence backing a mostly Neolithic origin of current populations.
Unsupervised admixture I believe tends to allocate ancient clinal mixed populations to their own component in the absence of "unadmixed" populations, peaking in the less admixed ones. It produces artificial results, that though informative about population relationships, are not as informative about actual components corresponding to actual ancient populations that underwent admixture.
For instance, these runs in Europe tend to produce an apparently homogeneous "North European" component modal in Lithuania, as well as a similar "South European" component modal in Sardinia. I've always suspected this didn't make any sense in relation to what we know of actual history, since if North European populations are of local Paleolithic continuity, they are very old, and would not coalesce into a single component spanning Iberia to Russia. Also hardly would any Neolithic component be concentrated in the North, since it would have to make it's way from further south. There are no inventions of agriculture recorded in Northern Europe, and the region has been well investigated by field researchers.
Sardinia is another problem. If this component was Paleolithic, why would it be present in so many countries next to other less related components? If it's Neolithic why would it not have absorbed some amount of other waves present in nearby populations?
These results cannot actually correspond to any population that ever existed, according to any historical, archaeological knowledge we have. Something was wrong. Thing is ADMIXTURE is a computer program and it doesn't know any history or archaeology. It can't test hypothesis by itself. Left alone it simply produces nice graphs of relatedness of populations, plus indications of admixture between more distant groups.

For more sensible results we have to direct the program according to historical data, and then see if hypothesis are validated. Also we can then apply our conclusions to new populations that may be available in the future.

No comments:

Post a Comment