Sunday, July 27, 2008

Structurama: Revisiting models for the inference of population structure

We've previously highlighted updates of Structure, commonly used for inferring population structure, and Distruct, the sister package used for pretty result plotting . Specifically, Structure uses a fixed number of populations (K) to assign posterior probabilities of assigning individuals to populations. It's worth pointing out that Huelsenbeck and Andolfatto (2007) have implemented a nice modification (originally suggested by Pella and Masuda 2006), which can consider K as a random variable that follows a Dirichlet process prior. In other words, it's no longer necessarily fixed to some arbitrary value. The Dirichlet process is nicely explained in the
Structurama manual.

In another homage to a previous post, it's also worth highlighting that it is remarkably easy to use R to manipulate the resulting plots (results are arrayed in columns) and create output very similar to that obtained in Distruct.

1 comment:

sergios-orestis kolokotronis said...

The famous Chinese restaurant process :-) It's even better to see John H describe said process at a conference like the 2006 Bayesian Invasion at Yale.