Post date: Dec 16, 2013 5:37:21 PM
We and predict survival on Ms for GLA when dropping sets of 20% of individuals (at least to a reasonable extent). I want to know whether this is also true if we drop 20% of families. This is a more difficult problem as we should have less information about the polygenic effect without siblings, but being able to predict survival without family members would allow us to rule out maternal effects as the drivers of our predictive ability.
Consequently, I created new phenotype files with family numbers (phenofam_* and phenotypeFamData.txt in melGemma) and created new training data sets by randomly dropping 20% of families (10 training sets each; pheno_trainFamSurv*, survival only). I am using these training sets to fit the BSLMM model (the probit model) and predict survival for the individuals (families) left out of the model fitting step. I am running two chains, with 5 million iterations, a 1 million iteration burnin and a thinning interval of 40. Thesea have job ids 67739-67818 on the dorc cluster.
These jobs sat in the long queue for 3 days. I decided to try them with a 96 hour walltime with the batch queue. The job numbers are 68320-68399 (submitted 19xii13 at 9:33 am).