Sunday, March 11, 2012

Experimental Design and Robustness

With the new release of Treemix, it is increasingly possible to attempt to map genetic drift and admixture within the human species.  If you read through Joe Pickrell and Jonathan Pritchard's paper, you will see buried deep in the Supplementary Material, a test for robustness of the allele data sets they have chosen.  It's a classic test of robustness against missing data.  In this test, they infer that there must be missing data, as the admixture between Oceanian populations with Neanderthals and Denisovans appears with one allele data set, but not the other.  The authors correctly reject any inferences that might be made for Oceanian populations with the given data sets.

With recent genetic discoveries, it is clear that low level admixture events with archaic hominids are present in most human populations, including Asians, Europeans, West Africans and Native Americans.  These are low level events, but as they lie at the root of human synthesis, genetic inferences which do not contain enough data to map these early admixture events are likely not to be robust.

Other model vulnerabilities such as a lack of robustness against model inadequacy may also be present.  Pickrell and Pritchard note that the Treemix algorithm depends on the assumption that human evolution is largely tree like.  Because of this, the algorithm has trouble handling early admixture events such as early human-Neanderthal admixture. 

Overall, Treemix appears to be a powerful tool, but it should definitely come with the label USER BEWARE.

Reference:

R. Srivastava, Rajender Parsad and V. K. Gupta;  Robustness Aspects of Experimental Designs, (Link)

No comments:

Post a Comment

Comments have temporarily been turned off. Because I currently have a heavy workload, I do not feel that I can do an acceptable job as moderator. Thanks for your understanding.

Note: Only a member of this blog may post a comment.