pifdr: utility for computing plug-in FDR

In gQTLstats: gQTLstats: computationally efficient analysis for eQTL and allied studies

Description

utility for computing plug-in FDR

Usage

pifdr( obs, perms, trimToUnit = TRUE, ... ) 
pifdr2( obs, perms, trimToUnit = TRUE, expandPerms=TRUE, ... ) 

Arguments

obs	observed association scores
perms	vector of association scores under permutation; length should be integer multiple of length(obs)
trimToUnit	logical, if TRUE, values greater than 1 are replaced by 1. Such values can occur, for example, with relatively small sample sizes.
expandPerms	With certain pair-specific filtering operations, the number of scores obtained after permutation may not be a multiple of the number of observed scores. If TRUE, the scores obtained under permutation are sampled with replacement to simplify computation of plug-in FDR.
...	extra arguments ignored

Details

Revised 12/30/13 to employ hist() to rapidly bin the permuted values.

Value

vector of plug-in FDR estimates congruent to obs

References

Hastie Tibshirani and Friedman Elements of Statistical Learning ch 18.7

Examples

set.seed(1234)
op = par(no.readonly=TRUE)
par(mfrow=c(2,2))
X = c(rchisq(30000,1),rchisq(300,10))
Y = rchisq(30300*3,1)
qqplot(Y, X, xlab="null", ylab="observed")
hist(pp <- pifdr(X,Y), xlab="plug-in FDR", main=" ")
library(multtest)
rawp = 1-pchisq(X, 1)
MT <- mt.rawp2adjp(rawp) 
MT2 = MT[[1]][order(MT[[2]]),]
plot(MT2[,"BH"], pp, xlab="BH FDR", ylab="plug-in FDR")
par(op)

gQTLstats documentation built on Nov. 8, 2020, 7:53 p.m.