Post date: Sep 02, 2013 10:49:15 PM
I ran estpost to calculate dic and summarize mcmc output for genotypes and fst. I automated this by creating a perl script for batch runs of estpost (on rc, ~/bin/batchestpost.pl), and submitted the batch runs to the rc cluster with a simple pbs script (sub.sh in /home/A01963476/scratch/admix/, which is where it runs from). I am still processing the genotypes, but I have summaries of dic and fst. Estimates of dic are remarkably consistent between chains. I am not very interested in selecting an appropriate number of clusters (populations), but based on these results k=4 best fits the common variant data and k=8 (or greater, this is the maximum I looked at) best fits the low frequency and rare variant data (dic plot). Estimates of fst are consistent between chains and vary little in magnitude (at least for k > 2) across k. But mean fst increased from rare to common variants (fst plot, again this is messier for k < 3). Thus, one interpretation of these results is that population structure is more pronounced for common variants, but more fine grain for rare variants.