gsri(exprs, groups, geneSet, names=NULL, weight=NULL, nBoot=100,
test=rowt, testArgs=NULL, alpha=0.05, grenander=TRUE, verbose=FALSE,
...)ExpressionSet containing
the expression intensities of the microarray. If a matrix the rows
represent the genes and the columns the samples, respectively.exprs has samples.GeneSet or
GeneSetCollection defining the gene set(s) used for the
analysis. If missing all genes of exprs are considered to be part
of the gene set. If an object of class GeneSet only these genes
are considered to be part of the gene set. If an object of class
GeneSetCollecton the analysis is performed for each gene set of
the collection individually.geneSet
argument. Has to have as many unique elements as gene sets in the
analysis.NULL all genes are
assumed to have the same weight. Please note that the weights are
defined in a relative way and thus any kind of positive
weights is feasable. Must have as many elements as eighter
the genes defined in geneSet or in exprs.groups argument. In this package, a t-test (rowt) and an F-test
(rowF) are already supplied, with rowt being the
default. Additionally, a custom test function can be used in order
to be able to include any feasible statistical test in the
analysis. For details, please see the ‘details’ section.test
function. For details, please see the ‘details’ section and the help
for test-functions.geneSet is an object of class
GeneSetCollection.minSize in exprs, the gene set is
ignored in the analysis.
GeneSetCollection. For details, please see the ‘details’
section.
Gsri with the slots:
result:cdf:parms:Gsri
class.
exprs part of the gene set:
signature(exprs="matrix", groups="factor", geneSet="missing")
signature(exprs="ExpressionSet", groups="factor", geneSet="missing")
GeneSet:
signature(exprs="matrix", groups="factor", geneSet="GeneSet")
signature(exprs="ExpressionSet", groups="factor", geneSet="GeneSet")
GeneSetCollection:
signature(exprs="matrix", groups="factor", geneSet="GeneSetCollection")
signature(exprs="ExpressionSet", groups="factor", geneSet="GeneSetCollection")
gsri method estimates the degree of differential expression in
gene sets. By assessing the part of the distribution of p-values
consistent with the null hypothesis the number of differentially
expressed genes is calculated.Through non-parametric fitting of the uniform component of the p-value distribution, the fraction of regulated genes $\sQuote{r}$ in a gene set is estimated. The GSRI $\sQuote{eta}$ is then defined as the $\sQuote{alpha*100}$%-quantile of the distribution of $\sQuote{r}$, obtained from bootstrapping the samples within the groups. The index indicates that with a probability of $(1-\sQuote{alpha})$% more than a fraction of $\sQuote{eta}$ genes in the gene set is differentially expressed. It can also be employed to test the hypothesis whether at least one gene in a gene set is regulated. Further, different gene sets can be compared or ranked according to the estimated amount of regulation.
Assessing the differential effect is based on p-values obtained from
statistical testing at the level of individual genes between the
groups. The GSRI approach is independent of the underlying test and
can be chosen according to the experimental design. With the t-test
(rowt) and F-test (rowF) two widely used statistical test are
already part of the package. Additional tests can easily used which
are passed with the test argument to the gsri method. For details
on how to implement custom test functions, please refer to the help of
rowt and rowF or the vignette of this package.
The GSRI approach further allows weighting the influence of individual genes in the estimation. This can be beneficial including for example the certainty that genes are part of a certain gene set derived from experimental findings or annotations.
Defining gene sets is available through the GSEABase package which
provides the GeneSet and GeneSetCollection classes a single or
multiple gene sets, respectively. This ensures a powerful approach for
obtaining gene sets from data objects, data bases, and other
bioconductor packages. For details on how to define or retrieve gene
sets, please refer to the documentation of the GSEABase package,
with a special focus on the GeneSet and GeneSetCollection classes.
The distribution of the p-values of a gene set is assessed in the cumulative density. In addition to a symmetrical empirical cumulative density function (ECDF), the modified Grenander estimator based on the assumption about the concave shape of the cumulative density is implemented and used by default. While the modified Grenander estimator reduces the variance and makes the approach more stable especially for small gene set, it underestimates the number of regulated genes and thus leads to conservative estimates.
In the case that the computation is performed for several gene sets in
the form of a GeneSetCollection object, it can be parallelized with the
multicore package. Please note that this package is not available
on all platforms. Using its capabilities requires attaching
multicore prior to the calculation and specification of the nCores
argument. For further details, please refer to the documentation of
the multicore package. This may be especially relevant in the case
that specific seed values for the bootstrapping are of interest.
GSRI-package
Class:
Gsri
Methods:
gsri
getGsri
getCdf
getParms
export
sortGsri
plot
show
summary
readCls
readGct
## Simulate expression data for a gene set of
## 100 genes, 20 samples (10 treatment, 10 control)
## and 30 regulated genes
set.seed(1)
exprs <- matrix(rnorm(100*20), 100)
exprs[1:30,1:10] <- rnorm(30*10, mean=2)
rownames(exprs) <- paste("g", 1:nrow(exprs), sep="")
groups <- factor(rep(1:2, each=10))
## Estimate the number of differentially expressed genes
res <- gsri(exprs, groups)
res
## Perform the analysis for different gene set
library(GSEABase)
gs1 <- GeneSet(paste("g", 25:40, sep=""), setName="set1")
gs2 <- GeneSet(paste("g", seq(1, nrow(exprs), by=5), sep=""), setName="set2")
gsc <- GeneSetCollection(gs1, gs2)
res2 <- gsri(exprs, groups, gs1)
res3 <- gsri(exprs, groups, gsc, verbose=TRUE)
summary(res2)
Run the code above in your browser using DataLab