Perplexing ADMIXTURE results - one component present at K =10 disappears at 11, reappears at 12

Question

[devenvyas]

Hello,

I looking at some ADMIXTURE analyses on human SNP data from the Affymetrix Human Origins array, and I am having a component disappear at K=11 and then re-appear at K=12. I have at least 10 replicates for every value of K for values 2 through 20.

CV error (K=1) 0.43206
CV error (K=2) 0.38759
CV error (K=3) 0.38116
CV error (K=4) 0.38019
CV error (K=5) 0.37783
CV error (K=6) 0.37654
CV error (K=7) 0.37589
CV error (K=8) 0.37577
CV error (K=9) 0.37559
CV error (K=10) 0.37556
CV error (K=11) 0.3762
CV error (K=12) 0.37605
CV error (K=13) 0.37639
CV error (K=14) 0.37705
CV error (K=15) 0.37766

I am looking K = 10, 11, and 12. While K=10 has the lowest CV error, a component of relevance to me doesn't show up until K=11. I've noticed that at K=10 (and lower too I believe), I have a component maximized in the Mbuti; however, at K=11 this component disappears and the Mbuti have proportions of different sub-Saharan African components and I have two new components arising (one maximized in the Mozabites and one in the Hazara). Then at K=12 the Mbuti component comes back and stays for good. This occurs across all replicates.

Has anyone else had similar ADMIXTURE results?

[Zev.Kronenberg]

Have you pruned linked sites?

[devenvyas]

Yup, I pruned the dataset (--indep-pairwise 200 25 0.4) based on the publication my comparative data came from. After posting this I was looking at a paper from some of my advisor's collaborators, and I saw similar things (http://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1004393 see figure S1 and look at Biaka from K=15-18 or Karretjie from 14-16).