Polyfit (version 1.6.0)

pfNbinomTest: The Polyfit extension to the DESeq functions nbinomTest() and nbinomTestForMatrices()

Description

Polyfit extensions to the DESeq functions nbinomTest and nbinomTestForMatrices which test for differences between the base means of two conditions (i.e., for differential expression in the case of RNA-Seq).

Usage

pfNbinomTest(cds, condA, condB, pvals_only = FALSE, eps = NULL)
pfNbinomTestForMatrices(countsA, countsB, sizeFactorsA, sizeFactorsB, dispsA, dispsB )

Arguments

cds

a CountDataSet with size factors and raw variance functions

condA

one of the conditions in 'cds'

condB

another one of the conditions in 'cds'

pvals_only

return only a vector of (unadjusted) p values instead of the data frame described below

eps

This argument is no longer used. Do not use it

countsA

A matrix of counts, where each column is a replicate

countsB

Another matrix of counts, where each column is a replicate

sizeFactorsA

Size factors for the columns of the matrix 'countsA'

sizeFactorsB

Size factors for the columns of the matrix 'countsB'

dispsA

The dispersions for 'countsA', a vector with one value per gene

dispsB

The same for 'countsB'

Value

pfNbinomTest gives a data frame with the following columns:

id

The ID of the observable, taken from the row names of the counts slots.

baseMean

The base mean (i.e., mean of the counts divided by the size factors) for the counts for both conditions

baseMeanA

The base mean (i.e., mean of the counts divided by the size factors) for the counts for condition A

baseMeanB

The base mean for condition B

foldChange

The ratio meanB/meanA

log2FoldChange

The log2 of the fold change

pval

The p value for rejecting the null hypothesis 'meanA==meanB'

padj

The adjusted p values (adjusted with 'p.adjust( pval, method="BH")')

pfNbinomTestForMatrices gives a vector of unadjusted p values, one for each row in the counts matrices.

Details

These functions have the same behaviour as the DESeq functions nbinomTest and nbinomTestForMatrices, except that the `flagpole' in the P-value histogram, particularly at p = 1 is redistributed using the function twoSidedPValueFromDiscrete.

References

Burden, C.J., Qureshi, S. and Wilson, S.R. (2014). Error estimates for the analysis of differential expression from RNA-seq count data, PeerJ PrePrints 2:e400v1.

Anders, S. and Huber, W. (2010). Differential expression analysis for sequence count data. Genome Biology, 11(10), R106.

Examples

cds <- makeExampleCountDataSet()
cds <- estimateSizeFactors( cds )
cds <- estimateDispersions( cds )
nbT <- nbinomTest( cds, "A", "B" )
head( nbT )
nbTPolyfit <- pfNbinomTest( cds, "A", "B" )
head( nbTPolyfit )

oldpar <- par(mfrow=c(1,2))
hist(nbT$pval,breaks=seq(0,1,by=0.01), 
   				xlab="P-value", main="DESeq")
hist(nbTPolyfit$pval,breaks=seq(0,1,by=0.01), 
 					xlab="P-value", main="polyfit-DESeq")
par(oldpar)