After DOD633, there are now 600 active and unrelated members in the Project. I have included all these members in a few experiments that were aimed to:
- Test the bootstrap-based standard error reporting in ADMIXTURE 1.12
- Create a simple global test that would allow me to get a quick feel of where a sample comes from
- Test a few of my observations about minute shifts towards distant populations that I have been writing about in my other blog of late.
- Compare supervised vs. unsupervised ADMIXTURE modes
I have included five ancestral groups in addition to the Project participants: Papuans, Karitiana Amerindians, Lithuanian/Tuscan Europeans, Mbuti/San Palaeoafricans, and She/Tujia East Asians. The analysis is based on 138,839 SNPs after quality-control and LD-based pruning.
There are four different experiments:
- Supervised ADMIXTURE analysis, with five ancestral groups (K=5)
- Unsupervised ADMIXTURE groups (K=6). Asian and European Caucasoids split at K=5 so I upped K to 6 in order for all the five ancestral groups to be recreated.
- Principal Components analysis (all samples)
- Principal Components analysis (Dodecad samples projected on 5 ancestral groups)
All samples (PC1: 4.05%, PC2: 2.70% of variance):
Projected (PC1: 12.73%, PC2: 7.85% of the variance):
The raw numbers for all participants can be found in the spreadsheet.