Post date: Dec 11, 2013 7:7:45 PM
I summarized the cross-validation prediction for survival from gemma. For now all I have done is made boxplots of the predicted phenotype (on the probit scale) for the left-out individuals for each cross-validation set and all 10 combined. The perl script for this is scripts/summarizePredict.pl. The four figures (population and plant combinations) are in melGemma/figs/ and are called survPred*.pdf. The results are pretty interesting. In GLA on Ms we a real (albeit far from perfect) ability to predict survival (median predictive value of survivors ~ 0.59 vs non-survivors 0.43). This is really remarkable and this experiment also has the highest pve. We also have a very slight predictive ability for GLA on Ac, but basically nothing for SLA on either host. I am not sure exactly what to think of all of this, but it is certainly cool that we have some ability to predict whether a larvae survives or dies (for GLA on Ms, its natal host) based on its genotype.