create_bum: Create a bum 'object'

Description Usage Arguments Details Value

View source: R/aheatmap.R

Description

Given a vector of p values, fit BUM and calculate FDR table

Usage

1
2
create_bum(pval, alphas = c(0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5),
  FDR = NULL, maskPthr = 1, pcutoff = 0.05)

Arguments

pval

pvalue vector

alphas

alpha values to create FDR table

FDR

FDR value to be used for gene selection; if NA, guess it from FDR table

maskPthr

when the P value contains a lot of 1's or close to 1, BUM model might fails. Thus we can mask out these P values as NA so that we get a good estimate and selection for significant events. Default is 1, which means no masking is done.

pcutoff

cutoff of p value to select indeces based on p values; default is 0.05

Details

Where the pval is highly significant, FDR tends to be small and the proportion of H0 genes is also small (pi0). It turns out the estimate of pi0 is still consistant (very similar to the qvalue package) and thus the Bum model fits really well even in this case. The problem arises when we want to achieve a FDR larger than that is possible. i.e. maximum FDR=0.2 but specified FDR=0.3, in which case, the cutoffSignificant() function will return a value larger than 1. This is not a problem as long as we truncate this cutoff to 1 (notice that even using qvalue package will not achieve FDR=0.3). Finally, Bum model is very consistent with the qvalue package.

Value

a list


nickytong/GenAnalysis documentation built on July 20, 2019, 8:57 a.m.