Thursday, May 31, 2018

Playing with Eigensoft qpAdm

I've been somewhat cautious in using qpAdm, because it looks much easier to use than it actually is.  You can make admixtures, be happy with them, but are your results as reliable as you think?   I am not going to bore you with details and instead make some examples.   Following three examples are made using same references (right file).


Results picked up from detail rows:

Secondly same results selected by the qpAdm:












The EHG stands for East European Hunter-Gatherer
The Belt_Cave Me stands for  Belt_Cave_Mesolithic_LC, the name was truncated during copy-paste operation. 

You can see that in the result of East European Hunter-Gatherer there is definitely something wrong, although the results picked up from detail rows look great. 

Update 1.6.2018 10:35

My answer to Anynyymi/Anonymous is:  yes it works at least for Latvians, but I won't continue making occasional tests with this left model, because I am more ambitious.  It is just not fair to ask for and show occasional results and let them become forgotten.  We need a larger demo.

1.089        0.779712     0.522     0.233     0.245







Update 1.6.2018 11:30

The example below shows the reason why I need more time and don't want to follow any trains of thought outside of my own head and make detailed tests.  Sorry, if I do, it could mislead readers and I just want to dedicate myself to the point here.



5 comments:

  1. Does that CWC Evenk Anatolia LatviaMN model work for Lithuanians or Ukrainians?

    ReplyDelete
    Replies
    1. It asks for testing. Although the right could work, and I am quite sure that it works for all western and central Eurasians, the left is to be found. We have no shortcuts, only hard work.

      Delete
    2. And of course, there is no right answers; CWC_Baltic + Latvian_MN + Anatolian Neolithic can be substituted by something else. Again, hard work.

      Delete
  2. Try keeping Evenk in the Latvian model, this might resolve the issue.

    ReplyDelete
    Replies
    1. Yes, but I am now modeling a larger bunch of Europeans. My goal is to get negligible standard error values rather than only small tail probs. Small jackknife standard error eliminates transitions between close ancestral groups.

      Delete

English preferred, because readers are international.

No more Anonymous posts.