Reference 3 Admixture K=4

UPDATE: With fixed Reference 3.

Continuing with the admixture analysis with our new reference 3 dataset.

Here's the results spreadsheet for K=4.

You can click on the legend to the right of the bar chart to sort by different ancestral components.

I have implemented something in this bar chart. When you click on the legend to sort by a specific component, it filters the results so that only those populations with at least 5% of that component are shown. Let me know what you think about this.

Fst divergences between estimated populations for K=4 in the form of an MDS plot.

And the numbers:
C1 C2 C3
C2 0.120
C3 0.158 0.198
C4 0.168 0.124 0.267


  1. Looking forward to the higher K results. I'm speculating the Onge will form a cluster with the Papuans.

    • they do have sharing with papuans (some 30-40%) but they cluster with Indians!

      • Since the Onge are so distinctive, it's possible they'll form their own ancestral component for a high enough value of K. But I expect them to have some affinity with Indians, Papuans and Southeast Asians.

        • I agree with you zack. But it will be interesting to see their sharing components with others, before they make their own cluster.

          • According to Reich et al, ASI were closest to Onge among current populations. So I expect Onge to be somewhat close to some of the high-ASI populations.

          • The problem with Reich was that they didn't analyze the Southeast Asians and that's why onges were skewed towards Indians!

  2. Read somewhere that Andaman islanders separated perhaps around 17kyr ago via a land-bridge in the ice age. In theory they should have closer affinity with Indians.

  3. I don't think so! Still Myanmar is not surveyed fully and also another branch i.e. M32 is not observed elsewhere, this theory would not hold!
    Moreover, the Y chromosomal hgp D is completely absent in India (except to some Northeast Indian populations who migrated from east).
    So the situation of Andamanese (mainly Onges, Jarawas and sentineles is not clear so far!

  4. I wonder why Oceanic and some South Asian populations have so much African noise here. Papuans are said to be genetically the most distant from Africans yet the algorithm can’t pick this up, at least not at the lower K’s.

    • Do you mean most distant or most easily distinguished?
      On 23andme when I look at the generic Nigerian person's global similarity, the Oceanians at 64.13% are the closest non-Africans to that Nigerian person.

    • Note that this small African percentage disappears once Admixture is run at higher K's where South Asian or Papuan components have appeared.

      I think this is just an artifact of Admixture trying to fit South Asians and Papuans into 3 ancestral populations.

  5. In the spreadsheet you have two populations labeled yukaghirs, but I can find only one in the chart. Did you combine them?


    • Fixed. Apparently, when I ask Google docs to replace a sheet, it doesn't actually replace a sheet and the last row was a holdover from before.

    • The population spreadsheet has dataset(it's part of 2 dataset) as part of the key , while the admixture results table doesn't distinguish between datasets (so essentially it's combined) .