compareAn: Comparison of IcaSet objects using correlation

In bitona/MineICA: Analysis of an ICA decomposition obtained on genomics data

Comparison of IcaSet objects using correlation

Description

Compare IcaSet objects by computing the correlation between either projection values of common features or genes, or contributions of common samples.

Usage

  compareAn(icaSets, labAn,
    type.corr = c("pearson", "spearman"), cutoff_zval = 0,
    level = c("samples", "features", "genes"))

Arguments

icaSets	list of IcaSet objects, e.g results of ICA decompositions obtained on several datasets.
labAn	vector of names for each icaSet, e.g the the names of the datasets on which were calculated the decompositions.
type.corr	Type of correlation to compute, either 'pearson' or 'spearman'.
cutoff_zval	either NULL or 0 (default) if all genes are used to compute the correlation between the components, or a threshold to compute the correlation on the genes that have at least a scaled projection higher than cutoff_zval. Will be used only when correlations are calculated on S or SByGene.
level	Data level of the IcaSet objects on which is applied the correlation. It must correspond to a feature shared by the IcaSet objects: 'samples' if they were applied to common samples (correlations are computed between matrix A), 'features' if they were applied to common features (correlations are computed between matrix S), 'genes' if they share gene IDs after annotation into genes (correlations are computed between matrix SByGene).

Details

The user must carefully choose the object on which the correlation will be computed. If level='samples', the correlations are based on the mixing matrices of the ICA decompositions (of dimension samples x components). 'A' will be typically chosen when the ICA decompositions were computed on the same dataset, or on datasets that include the same samples. If level='features' is chosen, the correlation is calculated between the source matrices (of dimension features x components) of the ICA decompositions. 'S' will be typically used when the ICA decompositions share common features (e.g same microarrays). If level='genes', the correlations are calculated on the attributes 'SByGene' which store the projections of the annotated features. 'SByGene' will be typically chosen when ICA were computed on datasets from different technologies, for which comparison is possible only after annotation into a common ID, like genes.

cutoff_zval is only used when level is one of c('genes','features'), in order to restrict the correlation to the contributing features or genes.

When cutoff_zval is specified, for each pair of components, genes or features that are included in the circle of center 0 and radius cutoff_zval are excluded from the computation of the correlation.

It must be taken into account by the user that if cutoff_zval is different from NULL or 0, the computation will be much slowler since each pair of component is treated individually.

Value

A list whose length equals the number of pairs of IcaSet and whose elements are outputs of function cor2An.

Author(s)

Anne Biton

See Also

cor2An

Examples

dat1 <- data.frame(matrix(rnorm(10000),ncol=10,nrow=1000))
rownames(dat1) <- paste("g", 1:1000, sep="")
colnames(dat1) <- paste("s", 1:10, sep="")
dat2 <- data.frame(matrix(rnorm(10000),ncol=10,nrow=1000))
rownames(dat2) <- paste("g", 1:1000, sep="")
colnames(dat2) <- paste("s", 1:10, sep="")

## run ICA
resJade1 <- runICA(X=dat1, nbComp=3, method = "JADE")
resJade2 <- runICA(X=dat2, nbComp=3, method = "JADE")

## build params
params <- buildMineICAParams(resPath="toy/")

## build IcaSet object
icaSettoy1 <- buildIcaSet(params=params, A=data.frame(resJade1$A), S=data.frame(resJade1$S),
                          dat=dat1, alreadyAnnot=TRUE)$icaSet
icaSettoy2 <- buildIcaSet(params=params, A=data.frame(resJade2$A), S=data.frame(resJade2$S),
                          dat=dat2, alreadyAnnot=TRUE)$icaSet

listPairCor <- compareAn(icaSets=list(icaSettoy1,icaSettoy2), labAn=c("toy1","toy2"),
                         type.corr="pearson", level="genes", cutoff_zval=0)


## Not run: 
#### Comparison of 2 ICA decompositions obtained on 2 different gene expression datasets.
## load the two datasets
library(breastCancerMAINZ)
library(breastCancerVDX)
data(mainz)
data(vdx)

## Define a function used to build two examples of IcaSet objects
treat <- function(es, annot="hgu133a.db") {
   es <- selectFeatures_IQR(es,10000)
   exprs(es) <- t(apply(exprs(es),1,scale,scale=FALSE))
   colnames(exprs(es)) <- sampleNames(es)
   resJade <- runICA(X=exprs(es), nbComp=10, method = "JADE", maxit=10000)
   resBuild <- buildIcaSet(params=buildMineICAParams(), A=data.frame(resJade$A), S=data.frame(resJade$S),
                        dat=exprs(es), pData=pData(es), refSamples=character(0),
                        annotation=annot, typeID= typeIDmainz,
                        chipManu = "affymetrix", mart=mart)
   icaSet <- resBuild$icaSet
}
## Build the two IcaSet objects
icaSetMainz <- treat(mainz)
icaSetVdx <- treat(vdx)

## The pearson correlation is used as a measure of association between the gene projections
# on the different components (type.corr="pearson").
listPairCor <- compareAn(icaSets=list(icaSetMainz,icaSetVdx),
labAn=c("Mainz","Vdx"), type.corr="pearson", level="genes", cutoff_zval=0)

## Same thing but adding a selection of genes on which the correlation between two components is computed:
# when considering pairs of components, only projections whose scaled values are not located within
# the circle of radius 1 are used to compute the correlation (cutoff_zval=1).
listPairCor <-  compareAn(icaSets=list(icaSetMainz,icaSetVdx),
labAn=c("Mainz","Vdx"), type.corr="pearson", cutoff_zval=1, level="genes")

## End(Not run)

bitona/MineICA documentation built on April 23, 2023, 1:41 p.m.