Post date: Aug 23, 2013 3:50:31 PM
I am now pretty confident that the entropy software is working properly. I did not find any bugs with Fst and I suspect the high values of Fst are real and caused by having more divergent populations and only common variants. It will be interesting to see how the results differ for low frequency and rare variants. Consequently, I began the 'real' analysis of common variants with entropy. I am running two chains for k=2..8 on the research computing cluster. These were sent to the long queue with a request for 336 hours of wall time, 8 gb of RAM, and no request for a specific executing cluster. All jobs are currently running (ids 5687-5700). I am only using the admixture proportion model (q not Q mat) and I am using the starting values from LDA with -s 50. The chains are running for 15,000 iterations with a 5000 iteration burnin and thinning interval of 5.