Post date: Aug 25, 2014 10:23:53 PM
The gemma probit models runs finished. The posterior distribution for PVE is approximately U(0,1), that is to say, it is the same as the prior distribution. Perhaps the run was simply too short, though this seems unlikely (mixing appeared fine). But, just in case I am now trying a run with 3 chains, 6 million iterations, and a 3 million iteration burnin:
gemma -g geno_int_timemaC.txt -p pheno_samplesC.txt -a annotation.txt -bslmm 3 -n 1 -o gemoutProbit_int_timemaCrep2 -rpace 40 -w 3000000 -s 6000000
Alternatively, variation in coverage could be contributing too much to the mean genotypes (average coverage is likely really low). So, I wrote a simple script to round the mean genotypes to the nearest integer and output geno_int* files that should mostly deal with this issue. I am running 3 additional chains under the same conditions with these input files:
gemma -g geno_int_timemaC.txt -p pheno_samplesC.txt -a annotation.txt -bslmm 3 -n 1 -o gemoutProbit_int_timemaCrep2 -rpace 40 -w 3000000 -s 6000000