calculateEFDR: Calculates the empricial false discovery rate.

In PeteHaitch/RUVcorr: Removal of unwanted variation for gene-gene correlations and related analysis.

Description

calculateEFDR returns the empirical false dicovery rate (EDFR) for supplied thresholds. This function also fits a loess curve to the estimated points. This allows the calculation of a threshold for priortisation of genes.

Usage

calculateEFDR(X, exclude, index.ref, set.size = length(index.ref), Weights,
  thresholds = seq(0.05, 1, 0.05), anno, Factor)

Arguments

X	A matrix of gene expression values.
exclude	A vector of indices of genes to exclude.
index.ref	A vector of indices of reference genes used for prioritisation.
set.size	A interger giving the size of the set of genes that are to be prioritised.
Weights	A object of class Weights or a list of weights. The weights should correspond to Factor. If NULL the unweighted correlations are used.
thresholds	A vector of thresholds; values should be in the range [0,1].
anno	A dataframe or a matrix containing the annotation of arrays in X.
Factor	A character string corresponding to a column name of anno.

Details

The empirical false discovery rate is estimated by drawing 1000 random sets of genes and calculating how many would be prioritised at every given threshold. A gene is is prioritised if at least one correlation with a known reference gene is above the given threshold.

Value

calculateEFDR returns an object of class EFDR. An object of class EFDR is a list with the following components:

• EFDR.values A vector of EDFRs.

• Thresholds A vector containing the values in threshold.

• loess.estimate An object of class loess.

Author(s)

Saskia Freytag

Examples

Y<-simulateGEdata(500, 500, 10, 2, 5, g=NULL, Sigma.eps=0.1, 250, 100, check.input=FALSE)
anno<-as.matrix(sample(1:4, dim(Y$Y)[1], replace=TRUE))
colnames(anno)<-"Factor"
weights<-findWeights(Y$Y, anno, "Factor")
calculateEFDR(Y$Y, exclude=251:500, index.ref=1:10, Weights=weights, anno=anno, Factor="Factor")

PeteHaitch/RUVcorr documentation built on May 8, 2019, 1:31 a.m.