Wednesday, July 9, 2014

robustness 20140709

gompShape batch regression




##############################
# batch regression ~ CV
pTb = 1: length(fit3[1,])
names(pTb) = names(fit3); length(pTb)
pTb[1:15]
#for( j in c(7:50) ) {
for( j in c(7:19, 22:32, 34:538) ) {
  m = lm( fit3[, j] ~ fit3$StddevLS/fit3$avgLS )
  sm = summary(m); sm
  pTb[j] = 1 - pf(sm$fsta[1], sm$fsta[2], sm$fsta[3])
}
pTb[pTb<0.05]
p= pTb[c(22:32, 34:538)]
p[p<0.01]
hist(p, br=20)
source("robust-fdr.R")
ret= robust.fdr(p)
summary(ret$q)
hist(ret$q, br=20)

#There are dozens of factors ~ rls CV, including the PPI degree. 




RLS CV ~ morphology CV (why a negative correlation?)



gompShape ~ morphology CV(positive correlation)?
Todo: I can exclude the small sample size measure to see if p-value can be improved.


RLS CV ~ (-)~ evolutionary measures. High selection pressure -> stronger robustness. 
    Kdata = read.csv( "data/Sce.Spa.KaKs.csv");

20140710note: The gene with small CV and high omega is highly interesting. 







Todo: use log, sqrt() transformation on variables. Remove small sample sized measures, say n<30?

No comments:

Post a Comment