After installing MixMapper (http://groups.csail.mit.edu/cb/mixmapper/) I found that I should purchase a full MATLAB license (200 Euros) to get working three-way mixture graphics. So I decided to move to Treemix. But before going ahead with Treemix I made a MixMapper run using Lazaridis’ data with additional later distributed ancient genomes. The idea was to build a basic construction using as many ancient genomes as possible and fit present-day populations on this result. I selected all ancient genomes reaching a 200 kSNP common coverage. Here is the result, a two-way mixture tree (click to enlarge):
Using this ancient structure as unmixed ancestry I ran MixMapper for north and west Europeans. It is worth noticing that many present-day populations get several possible mixtures, obviously because the genetic distance between ancient and present-day people is considerably big. Despite of that many results are very close what we see when analyzing modern populations, for example the Finns get around 93-98% European and the rest Karitiana. Karitiana and Mbuti are only modern groups included to the ancient sample set and without Karitiana the Finns would get MA1, Saqqaq or totally different ancient composition, likely with poorer fitting mixtures.
I got a hint that She samples from East Asia would be better than Karitiana, so I made this change. It looks like the She people are more distant to MA1 than Karitianas and MA1 moves to the ancestral line of Europeans, otherwise changes are small in Northeast Europe where this kind of admixture exists.
After checking the quality of ancient genomes I became sceptical. It looks like the quality of those genomes varies too much to get trustworthy results in wide genome comparisons. So my advice is to take these results with a grain of salt.