Package 'GiANT' reference manual

Title:	Gene Set Uncertainty in Enrichment Analysis
Description:	Toolbox for various enrichment analysis methods and quantification of uncertainty of gene sets, Schmid et al. (2016) <doi:10.1093/bioinformatics/btw030>.
Authors:	Florian Schmid [aut], Christoph Muessel [aut], Johann M. Kraus [aut], Hans A. Kestler [aut], Hans A. Kestler [cre]
Maintainer:	Hans A. Kestler <[email protected]>
License:	Artistic-2.0
Version:	1.3.4
Built:	2025-03-01 05:38:05 UTC
Source:	https://github.com/cran/GiANT

Enrichment analysis

Description

Toolbox for gene set analysis of uncertain gene sets.

Details

Package:	GiANT
Type:	Package
Version:	1.3
Date:	2020-04-29
License:	Artistic-2.0
LazyLoad:	yes

This package provides an approach for evaluating the fuzziness of a gene set. This is done by repeatedly performing gene set analyses on slightly modified versions of the gene set and comparing their enrichment scores. A utility for such uncertainty tests is provided in the evaluateGeneSetUncertainty function.

The package also comprises a generic framework for different types of enrichment analyses (Ackermann and Strimmer). It establishes a customizeable pipeline that typically consists of the following steps:

Calculation of gene-level statistics: A gene-level statistic scores the relationship between the measurements for a specific gene and the class labels. Typical measures include correlation coefficients, the t statistic or the fold change between the groups (see gls for gene-level statistics included in the package).
Transformation of gene-level statistic values: Optionally, the gene-level statistic values can be postprocessed, e.g. by taking the absolute value or the square for correlation values or by binarizing or ranking values. See transformation for transformations included in the package.
Calculation of gene set statistics: Based on the (possibly transformed) gene-level statistics, the gene set(s) of interest is/are scored. Examples are the median, the mean or the enrichment score of the gene-level statistic values in the gene set(s). See gss for gene set statistics included in the package.
Significance assessment: To assess the significance of the gene set statistic value(s) with respect to a null distribution, computer-intensive tests are performed. These tests repeatedly sample random label vectors or gene sets and calculate their gene set statistic values. These values can then be compared to the true gene set statistics. See significance for significance assessment methods included in the package.

The package represents such analysis pipelines as configuration objects that can be created using the function gsAnalysis. For predefined state-of-the-art methods, such as Gene Set Enrichment Analysis (Subramanian et al), Overrepresentation Analysis or Global Ancova (Hummel et al), it provides predefined configurations (see predefinedAnalyses).

The main function for standard gene set analyses, geneSetAnalysis, performs enrichment analyses based on pipeline configuration objects.

Author(s)

Florian Schmid, Christoph Müssel, Johann M. Kraus, Hans A. Kestler

Maintainer: Hans A. Kestler <[email protected]>

References

Ackermann, M., Strimmer, K. (2009) A general modular framework for gene set enrichment analysis. BMC Bioinformatics, 10(1), 47.

Subramanian, A., Tamayo, P., Mootha, V. K., Mukherjee, S., Ebert, B. L., Gillette, M. A., Paulovich, A., Pomeroy, S. L., Golub, T. R., Lander, E. S., Mesirov, J. P. (2005) Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proceedings of the National Academy of Science of the United States of America, 102, 15545-15550.

Hummel, M., Meister, R., Mansmann, U. (2008) Globalancova: exploration and assessment of gene group effects. Bioinformatics, 24(1), 78–85.

Examples


  data(exampleData)
  ##################################
  # Example 1: gene set analysis   #
  ##################################
  res <- geneSetAnalysis(  
    # parameters for geneSetAnalysis
    dat = countdata,
    geneSets = pathways[1],
    analysis = analysis.averageCorrelation(),
    adjustmentMethod = "fdr",
    # additional parameters for analysis.averageCorrelation
    labs = labels,
    method = "pearson",
    numSamples = 50)
  
  summary(res, mode="table")
  
  ####################################
  # Example 2: uncertainty analysis  #
  ####################################
  resUncertainty <- evaluateGeneSetUncertainty(  
    # parameters for evaluateGeneSetUncertainty
    dat = countdata,
    geneSet = pathways[[3]],
    analysis = analysis.averageCorrelation(),
    numSamplesUncertainty = 5,
    blockSize = 1,
    k = seq(0.1,0.9,by=0.1),
    # additional parameters for analysis.averageCorrelation
    labs = labels,
    numSamples = 5)
  
  plot(resUncertainty, main = names(pathways[3]))

data(exampleData)
  ##################################
  # Example 1: gene set analysis   #
  ##################################
  res <- geneSetAnalysis(  
    # parameters for geneSetAnalysis
    dat = countdata,
    geneSets = pathways[1],
    analysis = analysis.averageCorrelation(),
    adjustmentMethod = "fdr",
    # additional parameters for analysis.averageCorrelation
    labs = labels,
    method = "pearson",
    numSamples = 50)
  
  summary(res, mode="table")
  
  ####################################
  # Example 2: uncertainty analysis  #
  ####################################
  resUncertainty <- evaluateGeneSetUncertainty(  
    # parameters for evaluateGeneSetUncertainty
    dat = countdata,
    geneSet = pathways[[3]],
    analysis = analysis.averageCorrelation(),
    numSamplesUncertainty = 5,
    blockSize = 1,
    k = seq(0.1,0.9,by=0.1),
    # additional parameters for analysis.averageCorrelation
    labs = labels,
    numSamples = 5)
  
  plot(resUncertainty, main = names(pathways[3]))

Example data

Description

Randomly generated example data representing a count matrix

Usage

data(exampleData)data(exampleData)

Details

The data consists of a variable countdata with a matrix with 1113 genes and 96 samples. The count values are randomly generated numbers.

Examples

data(exampleData)

print(countdata)

data(exampleData)

print(countdata)

Create an overview table for an analysis

Description

Creates a data frame summarizing an analysis. This table has one row per gene set, each comprising the adjusted and unadjusted p-values and the number of genes for the set.

Usage

createSummaryTable(object,
		orderBy = c("adjustedPValues", "rawPValues", "geneSetName"),
		significantOnly = FALSE,
		signLevel = object$signLevel)
createSummaryTable(object,
		orderBy = c("adjustedPValues", "rawPValues", "geneSetName"),
		significantOnly = FALSE,
		signLevel = object$signLevel)

Arguments

`object`	A result object as returned by `geneSetAnalysis`.
`orderBy`	Specifies which field should be used for the row ordering. By default, rows are ordered according to the adjusted p-values.
`significantOnly`	Specifies whether all gene sets (`significantOnly=FALSE`) or only the statistically significant gene sets (`significantOnly=TRUE`) should be included in the table.
`signLevel`	If `significantOnly=TRUE`, this specifies the significance level for the results that should be included in the table. By default, the original significance level of the analysis is used.

Value

A data frame with one row for each included gene set and the columns "adjustedPValues", "rawPValues", "geneSetName" and "geneSetSize". For overrepresentation analyses, there is an additional column "intersectSize" specifying the size of the intersection of the core set and the corresponding gene set.

Examples

# load data
data(exampleData)
  # perform gene set analyses for several pathways
  res <- geneSetAnalysis(
  	# global parameters
	  dat = countdata,
	  geneSets = pathways, 
	  analysis = analysis.averageCorrelation(), 
	  # additional parameters for analysis.averageCorrelation
	  labs = labels,
	  numSamples = 10)

  tab <- createSummaryTable(res)
# load data
data(exampleData)
  # perform gene set analyses for several pathways
  res <- geneSetAnalysis(
  	# global parameters
	  dat = countdata,
	  geneSets = pathways, 
	  analysis = analysis.averageCorrelation(), 
	  # additional parameters for analysis.averageCorrelation
	  labs = labels,
	  numSamples = 10)

  tab <- createSummaryTable(res)

Quantify gene set uncertainty

Description

A robustness measure that quantifies the uncertainty of a gene set by performing a resampling experiment and can be used in the robustness parameter of gsAnalysis.

Usage

evaluateGeneSetUncertainty(
	...,
	dat,
	geneSet,
	analysis,
	numSamplesUncertainty,
	blockSize = 1,
	k = seq(0.01, 0.99, by=0.01),
	signLevel = 0.05,
	preprocessGeneSet = FALSE,
	cluster = NULL)
evaluateGeneSetUncertainty(
	...,
	dat,
	geneSet,
	analysis,
	numSamplesUncertainty,
	blockSize = 1,
	k = seq(0.01, 0.99, by=0.01),
	signLevel = 0.05,
	preprocessGeneSet = FALSE,
	cluster = NULL)

Arguments

`...`	Additional parameters for the different steps of the analysis pipeline, depending on the concrete configuration supplied in `analysis`.
`dat`	A numeric matrix of gene expression values for all analyzed genes. Here, each row corresponds to one gene, and each column corresponds to one sample. The rows must be named with the gene names used in the gene sets.
`geneSet`	A vector containing the names of genes in a gene set. All genes set must correspond to the row names of `dat`.
`analysis`	The parameters of the analysis that is applied to the perturbed copies of the gene set. These parameters are described by an object of class `gsAnalysis` as returned by the function `gsAnalysis` or the predefined analysis descriptors in `predefinedAnalyses`.
`numSamplesUncertainty`	The number of resampling experiments which should be applied to estimate the robustness of `geneSet`.
`blockSize`	Number of genes in one resampled block.
`k`	A `vector` of percentages of genes in the randomized gene sets that should be taken from the original gene set. The remaining genes are chosen randomly. For each value a resampling experiment is performed.
`signLevel`	The significance level for the significance assessment of the gene sets (defaults to `0.05`).
`preprocessGeneSet`	Specifies whether the gene sets in `geneSets` should be preprocessed or not. If set to `TRUE`, all genes that are not part of the data set (i.e. not in `rownames(dat)`) are removed from the gene sets.
`cluster`	If the analyses should be applied in parallel for the different values of `k`, this parameter must hold an initialized cluster as returned by `makeCluster`. If this parameter is `NULL`, the analyses are performed sequentially.

Details

The uncertainty analysis repeatedly replaces parts of the original gene sets by random genes and calculating the gene set statistics for these randomized gene sets. This yields a distribution of gene set statistic values for slightly modified variants of the original gene set.

Value

Returns a list (of class uncertaintyResult) with the following elements:

`uncertainty`	The calculated stability of the original gene set.
`confidenceValues`	A matrix of quantiles of `gssValues` (signLevel, 0.5, 1-signLevel). One row for each value in `k`.
`uncertaintyEvaluations`	A list with one entry per value in `k` containing the following elements: Quantiles of `gssValues`: signLevel, 0.5, 1-signLevel. gssValues: A vector of gene set statistic values, one for each randomly sampled gene set. uncertainGeneSets: A matrix containing all partially random gene sets. k: The percentage of genes in the randomized gene sets taken from the original gene set.
`signLevel`	The significance level used for this analysis.
`originalGeneSetValues`	Result of `geneSetAnalysis` for the original `geneSet`.

Examples

data(exampleData)

  res <- evaluateGeneSetUncertainty(
  	# parameters for evaluateGeneSetUncertainty
  	dat = countdata,
  	geneSet = pathways[[1]],
  	analysis = analysis.averageCorrelation(),
  	numSamplesUncertainty = 10,
  	k = seq(0.1,0.9, by=0.1),
  	# additional parameters for analysis.averageCorrelation
  	labs = labels,
  	numSamples = 10)
data(exampleData)

  res <- evaluateGeneSetUncertainty(
  	# parameters for evaluateGeneSetUncertainty
  	dat = countdata,
  	geneSet = pathways[[1]],
  	analysis = analysis.averageCorrelation(),
  	numSamplesUncertainty = 10,
  	k = seq(0.1,0.9, by=0.1),
  	# additional parameters for analysis.averageCorrelation
  	labs = labels,
  	numSamples = 10)

Filtering of gene sets

Description

Filters gene sets according to different criteria.

Usage

filterGeneSets(
	geneSets,
	includedGenes = NULL,
	minIntersectSize = length(includedGenes),
	adjMatrix = NULL,
	steps = NULL)
filterGeneSets(
	geneSets,
	includedGenes = NULL,
	minIntersectSize = length(includedGenes),
	adjMatrix = NULL,
	steps = NULL)

Arguments

`geneSets`	A list of gene sets, where each gene set is a vector of gene names corresponding to the row names of `dat`.
`includedGenes`	A vector of gene names whose occurence in each of the gene sets is checked. The further parameters how these genes are used to filter gene sets
`minIntersectSize`	If this parameter is not `NULL`, only gene sets with an intersection of at least `minIntersectSize` genes with respect to `includedGenes` (or `includedGenes` expanded by its interactions if `adjMatrix` and `steps` are supplied) are included in the result set. By default, this is the size of `includedGenes`, requiring `includedGenes` to be a subset of each gene set.
`adjMatrix`	An optional adjacency matrix in which an entry is `1` if there is a direct interaction between the corresponding genes and `0` otherwise. If this is non-null, the set of genes in `includedGenes` is expanded by adding all genes whose distance in the adjacency graph is at most `steps`.
`steps`	The maximum distance of interacting genes to the genes in `includedGenes` according to `adjMatrix` to be added to the expanded gene list. E.g., `steps = 1` means that all genes which are direct interaction partners of the initial genes in `includedGenes` are added to `includedGenes`.

Value

Returns a filtered list of gene sets with the same structure as geneSets.

Examples

geneSets <- list(
		gs1 = paste("gene",1:20,sep=""),
		gs2 = paste("gene",50:60,sep=""),
		gs3 = paste("gene",90:92,sep=""),
		gs4 = paste("gene",55:65,sep="")
	)

newGeneSets1 <- filterGeneSets(
	geneSets = geneSets,
	includedGenes = c("gene55","gene60"))

newGeneSets2 <- filterGeneSets(
	geneSets = geneSets,
	includedGenes = c("gene1","gene55","gene20","gene100"),
	minIntersectSize = 2)

examplePathway <- c("gene1","gene2","gene3","gene4")
pathwayAdjMatrix <- matrix(0,100,100)
rownames(pathwayAdjMatrix) <- paste("gene",1:100,sep="")
colnames(pathwayAdjMatrix) <- paste("gene",1:100,sep="")

# gene1 interacts with gene2 and gene3
# -> step 1 if gene1 is starting point
pathwayAdjMatrix[1,2:3] <- 1
pathwayAdjMatrix[2:3,1] <- 1

# gene3 interacts with gene4
# -> step 2 if gene1 is starting point
pathwayAdjMatrix[3,4] <- 1
pathwayAdjMatrix[4,3] <- 1

newGeneSets3 <- filterGeneSets(
	geneSets = geneSets,
	includedGenes = "gene1",
	adjMatrix = pathwayAdjMatrix,
	steps = 2)
geneSets <- list(
		gs1 = paste("gene",1:20,sep=""),
		gs2 = paste("gene",50:60,sep=""),
		gs3 = paste("gene",90:92,sep=""),
		gs4 = paste("gene",55:65,sep="")
	)

newGeneSets1 <- filterGeneSets(
	geneSets = geneSets,
	includedGenes = c("gene55","gene60"))

newGeneSets2 <- filterGeneSets(
	geneSets = geneSets,
	includedGenes = c("gene1","gene55","gene20","gene100"),
	minIntersectSize = 2)

examplePathway <- c("gene1","gene2","gene3","gene4")
pathwayAdjMatrix <- matrix(0,100,100)
rownames(pathwayAdjMatrix) <- paste("gene",1:100,sep="")
colnames(pathwayAdjMatrix) <- paste("gene",1:100,sep="")

# gene1 interacts with gene2 and gene3
# -> step 1 if gene1 is starting point
pathwayAdjMatrix[1,2:3] <- 1
pathwayAdjMatrix[2:3,1] <- 1

# gene3 interacts with gene4
# -> step 2 if gene1 is starting point
pathwayAdjMatrix[3,4] <- 1
pathwayAdjMatrix[4,3] <- 1

newGeneSets3 <- filterGeneSets(
	geneSets = geneSets,
	includedGenes = "gene1",
	adjMatrix = pathwayAdjMatrix,
	steps = 2)

Main interface for enrichment analyses.

Description

The main function of the package that performs a gene set analysis for a list of gene sets.

Usage

geneSetAnalysis(
	...,
	dat,
	geneSets,
	analysis,
	signLevel = 0.05,
	preprocessGeneSets = FALSE,
	adjustmentMethod = p.adjust.methods,
	cluster = NULL)
geneSetAnalysis(
	...,
	dat,
	geneSets,
	analysis,
	signLevel = 0.05,
	preprocessGeneSets = FALSE,
	adjustmentMethod = p.adjust.methods,
	cluster = NULL)

Arguments

`...`	Additional parameters for the different steps of the analysis pipeline, depending on the concrete configuration supplied in `analysis`.
`dat`	A numeric matrix of gene expression values for all analyzed genes. Here, each row corresponds to one gene, and each column corresponds to one sample. The rows must be named with the gene names used in the gene sets.
`geneSets`	A list of gene sets, where each gene set is a vector of gene names corresponding to the row names of `dat`.
`analysis`	An object of type `gsAnalysis` as returned by `gsAnalysis` or by the predefined configurations (see `predefinedAnalyses`).
`signLevel`	The significance level for the significance assessment of the gene sets (defaults to `0.05`).
`preprocessGeneSets`	Specifies whether the gene sets in `geneSets` should be preprocessed or not. If set to `TRUE`, all genes that are not part of the data set (i.e. not in `rownames(dat)`) are removed from the gene sets.
`adjustmentMethod`	The method to use for the adjustment for multiple testing (see `method` parameter of `p.adjust` for possible values).
`cluster`	If the analyses should be applied in parallel for the gene sets, this parameter must hold an initialized cluster as returned by `makeCluster`. If this parameter is `NULL`, the analyses are performed sequentially.

Details

This is the main interface function of the package for gene set enrichment analyses. Analyses usually consist of a pipeline of steps. Often, the first step is the calculation of a summary statistic for the relation of each gene to the class labels. These values or transformations thereof are employed to calculate a gene set statistic for each of the supplied gene sets. The significance of gene set enrichments can be determined according to different methods, and the robustness of gene sets can be evaluated by slightly modifying the gene sets. To provide a flexible mechanism for the plethora of different approaches arising from the different choices, basic pipeline configurations are encapsulated in gsAnalysis objects which can be created using the gsAnalysis function. Ready-to-use configuration objects for certain well-known methods are included in the package (see predefinedAnalyses). Parameters of the chosen analysis pipeline can be set in the ... parameter.

Value

An object of the type gsaResult with the following elements:

`adjustedPValues`	A vector of p-values, one for each gene set. These values are already adjusted for multiple testing according to the `adjustmentMethod` parameter.
`rawPValues`	The raw unadjusted p-values, one for each gene set.
`res.all`	A list comprising the detailed results for each gene set. Each element of this list is another list with the following components: `pValue`: The raw (unadjusted) p-value for the gene set. `geneSetValues`: If `analysis` is a global analysis, this is the object returned by the method for the corresponding gene set. For an analysis pipeline, this holds the values of the gene-level statistic, the transformed values and the values of the gene set statistic (see also `gsAnalysis`). `significanceValues`: Gene set statistics for each randomly drawn gene set for significance assessment and a `list` of this gene sets. Only set for analysis of type 'geneSetAnalysis'. `NULL` for 'global' analysis. `geneSet`: The supplied gene set.
`signLevel`	The significance level used for this analysis.
`analysis`	The performed analysis (of type `gsAnalysis`).
`analysisType`	A character string identifying the analysis as an enrichment analysis pipeline (`"geneSetAnalysis"`) or as a global analysis (`"global"`).
`adjustmentMethod`	The method used to adjust the p-values in `adjustedPValues`

References

Ackermann, M., Strimmer, K. (2009) A general modular framework for gene set enrichment analysis. BMC Bioinformatics, 10(1), 47.

Examples

# load data
data(exampleData)
  
  # apply predefined analysis for gene set enrichment analysis
  res <- geneSetAnalysis(	
  	# parameters for geneSetAnalysis
  	dat = countdata,
  	geneSets = pathways[1],
  	analysis = analysis.averageCorrelation(),
  	adjustmentMethod = "fdr",
  	# additional parameters for analysis.averageCorrelation
  	labs = labels,
  	method = "pearson",
  	numSamples = 10)
# load data
data(exampleData)
  
  # apply predefined analysis for gene set enrichment analysis
  res <- geneSetAnalysis(	
  	# parameters for geneSetAnalysis
  	dat = countdata,
  	geneSets = pathways[1],
  	analysis = analysis.averageCorrelation(),
  	adjustmentMethod = "fdr",
  	# additional parameters for analysis.averageCorrelation
  	labs = labels,
  	method = "pearson",
  	numSamples = 10)

Gene-level statistics

Description

Functions to calculate the gene-level statistic, as used in the gls parameter of gsAnalysis. A gene-level statistic calculates a measure of correlation between the expression of a gene and the class labels.

Usage

gls.cor(dat, labs, method = "pearson")

gls.regression(dat, labs)

gls.foldChange(dat, labs, logMeasurements = TRUE)

gls.tStatistic(dat, labs, pValue = FALSE, alternative = "two.sided")

gls.moderateTStatistic(dat,labs)

gls.nBinomTest(dat, labs,
	returnValue = c("pval", "qval", "foldChange", "log2FoldChange"),
	dispersionMethod = "blind",
	dispersionSharingMode = "fit-only",
	dispersionFitType = "local")
gls.cor(dat, labs, method = "pearson")

gls.regression(dat, labs)

gls.foldChange(dat, labs, logMeasurements = TRUE)

gls.tStatistic(dat, labs, pValue = FALSE, alternative = "two.sided")

gls.moderateTStatistic(dat,labs)

gls.nBinomTest(dat, labs,
	returnValue = c("pval", "qval", "foldChange", "log2FoldChange"),
	dispersionMethod = "blind",
	dispersionSharingMode = "fit-only",
	dispersionFitType = "local")

Arguments

`dat`	A numeric matrix of gene expression values for all analyzed genes. Here, each row corresponds to one gene, and each column corresponds to one sample. The rows must be named with the gene names used in the gene sets.
`labs`	A vector of class labels for the samples in `dat`.
`logMeasurements`	For `gls.foldChange`, whether the values in `dat` are logarithmized (`logMeasurements=TRUE`) or not (`logMeasurements=FALSE`).
`method`	For `gls.cor`, the correlation method to be used (see `cor`).
`pValue`	For `gls.tStatistic`, this specifies whether the p-value (`pValue=TRUE`) or the test statistic (`pValue=FALSE`) of the t test should be returned.
`alternative`	For `gls.tStatistic`, this specifies the alternative of the t-test. See also `t.test`.
`returnValue`	For `gls.nBinomTest`, this determines the type of values values that should be returned. `"pval"` returns the raw p-values, `"qval"` returns the p-values adjusted by the FDR, `"foldChange"` returns the fold changes, and `"log2FoldChange"` returns the log2 fold changes. For more details, see `results`.
`dispersionMethod`	For `gls.nBinomTest`, this specifies how the empirical dispersion is computed (see `estimateDispersions`).
`dispersionSharingMode`	For `gls.nBinomTest`, this specifies which values should be used by `results` (fitted values or empirical values, see `estimateDispersions` for more details).
`dispersionFitType`	For `gls.nBinomTest`, this determines the method for fitting the dispersion-mean relation (see `estimateDispersions`).

Details

Standard functions for the calculation of gene-level statistics (to be used in an analysis pipeline defined by gsAnalysis):

gls.cor: Calculates the correlation of the gene expression values to the class labels.
gls.regression: Calculates the slope of a linear regression of the gene expression values and the class labels.
gls.foldChange: Calculates the (standard or log2) fold change between the measurements for the two classes.
gls.tStatistic: Calculates the p-value or the statistic of a two-sample t test for the measurements of the two classes
gls.moderateTStatistic: Calculates the moderate t statistic for the measurements of the two classes
gls.nBinomTest: Applies the negative binomial test for sequencing data based on the DESeq2 package to test for differences between two classes (see results).

Value

Each of these function returns a numeric vector of gene-level statistics (one entry per gene).

Gene set statistics

Description

Functions to calculate a gene set statistic, as used in the gss parameter of gsAnalysis. A gene set statistic summarizes a single gene set.

Usage

gss.mean(x, geneSetIndices)

gss.sum(x, geneSetIndices)

gss.wilcoxonRankTest(x, geneSetIndices)

gss.maxmean(x, geneSetIndices)

gss.median(x, geneSetIndices)

gss.enrichmentScore(x, geneSetIndices, p = 1)

gss.fisherExactTest(x, geneSetIndices)

gss.gsz(x, geneSetIndices, w1 = 0.2, w2 = 0.5, preVar = 0, varConstant = 10)
gss.mean(x, geneSetIndices)

gss.sum(x, geneSetIndices)

gss.wilcoxonRankTest(x, geneSetIndices)

gss.maxmean(x, geneSetIndices)

gss.median(x, geneSetIndices)

gss.enrichmentScore(x, geneSetIndices, p = 1)

gss.fisherExactTest(x, geneSetIndices)

gss.gsz(x, geneSetIndices, w1 = 0.2, w2 = 0.5, preVar = 0, varConstant = 10)

Arguments

`x`	A vector comprising one numeric value for each gene in the data set. This vector is usually obtained from the previous step, the gene-level statistic (see `gls`) or the transformed gene-level statistic (see `transformation`).
`geneSetIndices`	A vector containing the indices of the genes in the gene set with respect to the full gene set (i.e., the indices of the rows containing the measurements for these genes in `dat`).
`p`	Factor for `gss.enrichmentScore` that specifies the way hits are weighted. For `p = 0`, the enrichment score is a Kolmogorov-Smirnov statistic. For `p = 1` (the default), hits are weighted by their correlation.
`w1`	Weight for the median of the variance estimates for a gene set of size `varConstant`. Should be between 0 and 1. Default is `w1 = 0.2`.
`w2`	Weight for the median of the variance estimates for a gene set across the whole gene list. Should be between 0 and 1. Default is `w2 = 0.5`.
`preVar`	Parameter for incorporating the uncertainty of the observations. This is omitted by default (`preVar = 0`).
`varConstant`	Reference gene set size used for variance estimates. Default is `varConstant = 10`.

Details

Standard functions for the calculation of gene set statistics (to be used in an analysis pipeline defined by gsAnalysis):

gss.mean: Calculates the mean of the (transformed) gene-level statistic values for the genes in the set.
gss.sum: Calculates the sum of the (transformed) gene-level statistic values for the genes in the set.
gss.wilcoxonRankTest: Calculates a Wilcoxon test comparing the (transformed) gene-level statistic values for the genes in the set versus those of the genes not in the set.
gss.maxmean: Calculates the maximum of the means of positive and negative statistic values, weighted by the overall proportion of positive/negative values (e.g. for correlation scores where the sign denotes the direction). Described in Efron and Tibshirani.
gss.median: Calculates the median of the (transformed) gene-level statistic values for the genes in the set.
gss.enrichmentScore: Calculates the enrichment score of the (transformed) gene-level statistic values for the genes in the set, as described in Subramanian et al.
gss.fisherExactTest: Performs Fisher's exact test to check gene sets for overrepresentation in the differential genes. This should be used used in combination with the transformation transformation.adjustAndBinarize.
gss.gsz: Calculates the Gene Set Z-score of the (transformed) gene-level statistic values for the genes in the set. Described in Toronen et al.

Value

Each method returns a single numeric value, the gene set statistic for the supplied gene set.

References

Efron, B., Tibshirani, R. (2007) On testing the significance of sets of genes. Annals of Applied Statistics, 1, 107-129.

Toronen, P., Ojala, P. J., Marttinen, P., Holm L. (2009) Robust extraction of functional signals from gene set analysis using a generalized threshold free scoring function. BMC Bioinformatics, 10(1), 307.

Global analyses

Description

Functions to perform global gene set analyses, as used in the globalStat parameter of gsAnalysis.

Usage

global.overrepresentation(dat,
	geneSet,
	coreSet)

global.ancova(dat,
	geneSet,
	labs,
	...)

global.test(dat,
	geneSet,
	labs,
	...)
global.overrepresentation(dat,
	geneSet,
	coreSet)

global.ancova(dat,
	geneSet,
	labs,
	...)

global.test(dat,
	geneSet,
	labs,
	...)

Arguments

`dat`	A numeric matrix of gene expression values for all analyzed genes. Here, each row corresponds to one gene, and each column corresponds to one sample. The rows must be named with the gene names used in the gene sets.
`geneSet`	A gene set in form of a vector of gene names corresponding to the row names of `dat`.
`coreSet`	A gene set of interest resulting from an analysis of `dat` that should be compared to `geneSet` in the overrepresentation analysis. This is also a vector of gene names corresponding to the row names of `dat`.
`labs`	A vector of class labels for the samples in `dat`.
`...`	Further parameters for `GlobalAncova` and `gt`, as defined in the corresponding manual pages. The parameters `xx`, `test.genes` and `group` are set automatically by `global.ancova`, and the parameters `alternative`, `subsets` and `response` are set automatically by `global.test`.

Details

Wrapper functions for global gene set analyses.

global.overrepresentation: This function performs an overrepresentation analysis by rating the overlap of geneSet and coreSet with respect to the set of all genes using Fisher's exact test.
global.ancova: This function performs a global gene set enrichment analysis using the global ANCOVA method by Hummel et al. It wraps the GlobalAncova function in the GlobalAncova package.
global.test: This function performs a global gene set enrichment analysis using a global test by Goeman et al. It wraps the gt function in the globaltest package.

Value

A list containing the following items:

`pValue`	The p-value for the significance of `geneSet`.
`intersectGeneSetCoreSet`	This element is only returned in case of an overrepresentation analysis and consists of a vector of genes included in both sets (`geneSet` and `coreSet`).
`res.all`	The full result object returned by `fisher.test`, `GlobalAncova` or `gt` respectively.

References

Hummel, M., Meister, R., Mansmann, U. (2008) GlobalANCOVA: exploration and assessment of gene group effects. Bioinformatics, 24(1), 78–85.

Goeman, J. J., van de Geer, S. A., de Kort, F., van Houwelingen, H. C. (2004) A global test for groups of genes: testing association with a clinical outcome. Bioinformatics, 20(1), 93–99.

Gene set analysis.

Description

Defines the configuration of an analysis that can be performed using geneSetAnalysis, and returns it as a wrapper object.

Usage

gsAnalysis(name,
	gls = NULL,
	glsParameterNames = NULL,
	transformation = NULL,
	transformationParameterNames = NULL,
	gss = NULL,
	gssParameterNames = NULL,
	globalStat = NULL,
	globalStatParameterNames = NULL,
	significance = NULL,
	significanceParameterNames = NULL,
	testAlternative = c("greater", "less"))

gsAnalysis(name,
	gls = NULL,
	glsParameterNames = NULL,
	transformation = NULL,
	transformationParameterNames = NULL,
	gss = NULL,
	gssParameterNames = NULL,
	globalStat = NULL,
	globalStatParameterNames = NULL,
	significance = NULL,
	significanceParameterNames = NULL,
	testAlternative = c("greater", "less"))

Arguments

`name`	A character string describing the analysis.
`gls`	The name of the function that calculates the gene-level statistic for a given dataset. If set to `NULL`, it is assumed that the input data already comprises gene-level statistic values, and the input is directly supplied to `transformation`. The first (fixed) parameter of a gls function is the dataset.
`glsParameterNames`	A character vector of names of the parameters used by the gene-level statistic defined in `gls`.
`transformation`	The name of the function that transforms the gene-level statistics values. If set to `NULL`, the values supplied by `gls` are directly handed over to `gss`. The only fixed parameter for transformation is the gene-level statistic (supplied as first parameter).
`transformationParameterNames`	A character vector of names of the parameters used by the transformation defined in `transformation`.
`gss`	The name of the function that calculates the gene set statistics from untransformed or transformed gene-level statistic values. If set to `NULL`, the values supplied by `transformation` are directly handed over to `significance`. Fixed parameters are the transformed values (first parameter) and `geneSetIndices` containing the (row-) indices of the current gene set genes in the dataset.
`gssParameterNames`	A character vector vector of names of the parameters used by the gene set statistic defined in `gss`.
`globalStat`	If the gene set analysis consists of a global test and cannot be divided into the calculation of a gene-level statistic, a transformation and the calculation of the gene set statistics, this parameter defines the name of the function that performs this global analysis. In this case, the parameters `gls`, `transformation` and `gss` are ignored. Fixed parameters are `dat` containing the whole dataset and `geneSet` containing the current gene set.
`globalStatParameterNames`	A character vector names of the parameters used by the global analysis defined in `globalStat`.
`significance`	The name of a method that performs a significance assessment for the gene set statistic values. If set to `NULL`, `geneSetAnalysis` does not return p-values, but returns the statistics supplied by `gss` or `globalStat`. Fixed parameters are `dat` containing the whole dataset, `geneSet` containing the current gene set, `analysis` with the supplied `gsAnalysis` and `glsValues` with (depending on whether a transformation is supplied or not) transformed gene-level statistics for each gene in the dataset.
`significanceParameterNames`	A character vector of names of the parameters used by the significance assessment method defined in `significance`.
`testAlternative`	Specifies the alternative hypothesis of the significance test for the gene set enrichment, which may be dependent on the chosen gene set statistic. Must be one of `"greater"` or `"less"`.

Details

The function provides a way of flexibly defining the steps of the gene set analysis pipeline. This pipeline consists of a subset of the following steps, each of which may have specific parameters:

Gene-level analysis: A gene-level statistic scores the relationship between the measurements for a specific gene and the class labels. Typical measures include correlation coefficients, the t statistic or the fold change between the groups (see gls for gene-level statistics included in the package).
Transformation of gene-level statistics: Optionally, the gene-level statistic values can be postprocessed, e.g. by taking the absolute value or the square for correlation values or by binarizing or ranking values. See transformation for transformations included in the package.
Gene set analysis: Based on the (possibly transformed) gene-level statistics, the gene set(s) of interest is/are scored. Examples are the median, the mean or the enrichment score of the gene-level statistic values in the gene set(s). See gss for gene set statistics included in the package.
Significance assessment: To assess the significance of the gene set statistic value(s) with respect to a null distribution, computer-intensive tests are performed. These tests repeatedly sample random label vectors or gene sets and calculate their gene set statistic values. These values can then be compared to the true gene set statistics. See significance for significance assessment methods included in the package.
Global analysis: As an alternative to the above pipeline steps, it is possible to define a single, global method that directly calculates an enrichment p-value for a supplied data set and gene set. See global for the global analysis tests included in the package.

Several state-of-the-art analyses have predefined configuration objects in which the above steps are defined accordingly (see predefinedAnalyses).

Value

An object of class gsAnalysis with components corresponding to the above parameters.

Examples

	data(exampleData)
# defines an analysis that corresponds to gsAna1()
gsa <- gsAnalysis(
	name = "averageCorrelation",
	gls = "gls.cor",
	glsParameterNames = c("labs","method"),
	transformation = "transformation.abs",
	transformationParameterNames = NULL,
	gss = "gss.mean",
	gssParameterNames = NULL,
	globalStat = NULL,
	globalStatParameterNames = NULL,
	significance = "significance.sampling",
	significanceParameterNames = c("numSamples"),
	testAlternative = "greater")
print(gsa)


  # apply the previously defined analysis
  res <- geneSetAnalysis(
  	# global parameters
  	dat = countdata,
  	geneSets = pathways[1], 
  	analysis = gsa, 
  	# parameters for the specific analysis gsAna1
  	labs = labels,
  	numSamples = 10)
data(exampleData)
# defines an analysis that corresponds to gsAna1()
gsa <- gsAnalysis(
	name = "averageCorrelation",
	gls = "gls.cor",
	glsParameterNames = c("labs","method"),
	transformation = "transformation.abs",
	transformationParameterNames = NULL,
	gss = "gss.mean",
	gssParameterNames = NULL,
	globalStat = NULL,
	globalStatParameterNames = NULL,
	significance = "significance.sampling",
	significanceParameterNames = c("numSamples"),
	testAlternative = "greater")
print(gsa)


  # apply the previously defined analysis
  res <- geneSetAnalysis(
  	# global parameters
  	dat = countdata,
  	geneSets = pathways[1], 
  	analysis = gsa, 
  	# parameters for the specific analysis gsAna1
  	labs = labels,
  	numSamples = 10)

Null distribution histogram and statistic of the input set for enrichment analyses.

Description

Plots the distribution of gene set statistic values obtained in different resampling settings of an enrichment analysis, and draws the statistic value of the input set as a vertical line.

Usage

## S3 method for class 'gsaResult'
hist(x,
	signLevel = x$signLevel,
	subset = NULL,
	ask = FALSE,
	addLegend = TRUE,
	...)
## S3 method for class 'gsaResult'
hist(x,
	signLevel = x$signLevel,
	subset = NULL,
	ask = FALSE,
	addLegend = TRUE,
	...)

Arguments

`x`	A result of a call to `geneSetAnalysis` (see also Details).
`signLevel`	The significance level that should be applied for the plots. Default is the significance level used for the analysis in `x`.
`subset`	Indices for the results that should be included in the diagram.
`ask`	If set to true, the plot function will prompt for a user input for each new plot that is shown on an interactive device (see `par("ask")`).
`addLegend`	If set to true (default), a `legend` is added to the plot.
`...`	Other parameters which can be used for histograms (see `hist`).

Details

The function plots the distribution of gene set statistic values under the null hypothesis. It requires the significance assessment step of the enrichment analysis configuration (parameter significance or gsAnalysis) to be a computer-intensive testing procedure that yields a distribution of gene set statistic p-values under the null hypothesis. Predefined configurations for which this plot works are analysis.gsea, analysis.averageCorrelation and analysis.averageTStatistic.

A histogram is plotted for the analysis in x. If x includes the analyses for several gene sets, one histogram is plotted for each of the gene sets.

The statistic value of the input set is depicted as a vertical line.

The most common graphical parameters can be supplied as vectors (one entry per analyzed gene set) to vary them between the different analyses. These parameters are: main, xlab, ylab.

Value

Returns a list with all the underlying data for the plotted histograms as invisible object.

Examples

# load data
data(exampleData)
  res <- geneSetAnalysis(
  	# global parameters
  	dat = countdata,
  	geneSets = pathways[3], 
  	analysis = analysis.averageCorrelation(), 
  	# additional parameters for analysis.averageCorrelation
  	labs = labels,
  	p = 1,
  	numSamples = 10)
  
  # plot the histogram for the cell cycle control gene set
  hist(res, main = names(pathways[3]))
# load data
data(exampleData)
  res <- geneSetAnalysis(
  	# global parameters
  	dat = countdata,
  	geneSets = pathways[3], 
  	analysis = analysis.averageCorrelation(), 
  	# additional parameters for analysis.averageCorrelation
  	labs = labels,
  	p = 1,
  	numSamples = 10)
  
  # plot the histogram for the cell cycle control gene set
  hist(res, main = names(pathways[3]))

Example data to label samples in countdata (rows) in two classes 0 and 1

Description

Randomly generated example data to give samples in countdata a class label 0 or 1

Usage

data(exampleData)data(exampleData)

Details

The data consists of a variable labels with a vector of 96 values. The values are a random sequence of 0's and 1's.

Examples

data(exampleData)

print(labels)

data(exampleData)

print(labels)

Merge multiple probes for one gene

Description

Merges all probes belonging to the same gene by identifying duplicate row names in a data matrix.

Usage

mergeProbesForGenes(dat,
	method = c("mean", "max", "min", "median"))
mergeProbesForGenes(dat,
	method = c("mean", "max", "min", "median"))

Arguments

`dat`	A numeric matrix of gene expression values for all analyzed genes. Here, each row corresponds to one probe, and each column corresponds to one sample. The rows must be named with the gene names and may contain duplicates if multiple probes correspond to the same gene.
`method`	The method which should be used to merge probes entries in `dat`. Depending on the chosen method, the merged value for a gene and a specific sample is defined as the mean value, the maximum value, the minimum value or the median of all probes of this sample belonging to the gene.

Value

A matrix of the same structure as dat, but possibly with fewer rows if probes were merged.

Examples

dat <- matrix(1:6, nrow=3, ncol=2)
rownames(dat) <- c("g1", "g2", "g1")

newDat <- mergeProbesForGenes(dat, method = "mean")
dat <- matrix(1:6, nrow=3, ncol=2)
rownames(dat) <- c("g1", "g2", "g1")

newDat <- mergeProbesForGenes(dat, method = "mean")

Venn Euler Diagramm

Description

Parses a GMT file as downloadable from MSigDB (presented in Subramanian et al.) and returns a list of gene sets.

Usage

	parseGmt(file)
parseGmt(file)

Arguments

file

A file name.

Details

Parses a GMT file and returns a list of gene sets. Each list element named according to the included gene set. The gene set files can be downloaded from http://www.broadinstitute.org/gsea/msigdb.

Value

A named list of gene sets.

References

Example data for pathways to to gene set enrichtment analyses

Description

A list comprising exemplary pathways. Each pathway is represented by a vectro of random genes from countdata which are associated to this pathway.

Usage

data(exampleData)data(exampleData)

Details

The data consists of a variable pathways of class list with 9 exemplary pathways. Each element of the list corresponds to one pathway - represented by a vector of gene names which are involved in this pathway.

Examples

data(exampleData)

print(pathways)

data(exampleData)

print(pathways)

Plots the results of an uncertainty analysis.

Description

For each percentage of original gene set genes, the quantiles of the distribution obtained by a resampling simulation are plotted. Significance threshold (quantile of the Null distribution) and the test statistic of the original gene set are drawn as horizontal lines.

Usage

## S3 method for class 'uncertaintyResult'
plot(x,
	signLevel = x$signLevel,
	addLegend = TRUE,
	addMinimalStability = FALSE,
	...)
## S3 method for class 'uncertaintyResult'
plot(x,
	signLevel = x$signLevel,
	addLegend = TRUE,
	addMinimalStability = FALSE,
	...)

Arguments

`x`	A result of a call to `evaluateGeneSetUncertainty` (see also Details).
`signLevel`	Only results with significance level smaller than the given value are plotted.
`addLegend`	If set to true (default), a `legend` is added to the plot.
`addMinimalStability`	If set to true, a line is added to the plot giving the minimal stability.
`...`	Other parameters which can be used for histograms (see `plot`).

Details

The function plots the quantiles of the resampling distributions for evaluated degrees of fuzziness. It requires the significance assessment step of the enrichment analysis configuration (parameter significance or gsAnalysis) to be a computer-intensive testing procedure that yields a distribution of gene set statistic values under the null hypothesis. Predefined configurations for which this plot works are analysis.gsea, analysis.averageCorrelation and analysis.averageTStatistic.

Three lines, corresponding to the different qunatiles with one dot per fuzziness evaluation (k) are plotted for the analysis in x. The significance threshold is shown as a green horizontal line. The statistic value of the original input set is depicted as a red horizontal line.

If addMinimalStability is TRUE, the lower bound of the stability is ploted as a dotted line.

Value

No return value, called for creating a plot.

Examples

# load data
data(exampleData)

  res <- evaluateGeneSetUncertainty(
  	# parameters for evaluateGeneSetUncertainty
  	dat = countdata,
  	geneSet = pathways[[1]],
  	analysis = analysis.averageCorrelation(),
  	numSamplesUncertainty = 10,
  	N = seq(0.1,0.9, by=0.1),
  	# additional parameters for analysis.averageCorrelation
  	labs = labels,
  	numSamples = 10)
  
  # plot the results for the cell cycle control gene set
  plot(res, addMinimalStability = TRUE)
# load data
data(exampleData)

  res <- evaluateGeneSetUncertainty(
  	# parameters for evaluateGeneSetUncertainty
  	dat = countdata,
  	geneSet = pathways[[1]],
  	analysis = analysis.averageCorrelation(),
  	numSamplesUncertainty = 10,
  	N = seq(0.1,0.9, by=0.1),
  	# additional parameters for analysis.averageCorrelation
  	labs = labels,
  	numSamples = 10)
  
  # plot the results for the cell cycle control gene set
  plot(res, addMinimalStability = TRUE)

Plot overlap of gene sets and core set

Description

Plots a Venn diagramm of the overlaps of the core set and gene sets in an overrepresentation analysis.

Usage

plotOverrepresentation(
	object,
	signLevel = object$signLevel,
	subset = NULL,
	aggregate = FALSE,
	ask = FALSE,
	...)
plotOverrepresentation(
	object,
	signLevel = object$signLevel,
	subset = NULL,
	aggregate = FALSE,
	ask = FALSE,
	...)

Arguments

`object`	A result of a call to `geneSetAnalysis` using the predefined analysis `analysis.customOverrepresentation` or `analysis.overrepresentation`.
`signLevel`	Only results with significance level smaller than the given value are included in the venn diagram.
`subset`	Indices for the results that should be included in the diagram.
`aggregate`	Specifies whether all gene sets should be plotted in a single Venn diagram (which is possible for at most four gene sets) or whether there should be one Venn diagram for each gene set.
`ask`	If set to true, the plot function will prompt for a user input for each new plot that is shown on an interactive device (see `par("ask")`). If `aggregate = TRUE`, `ask` is ignored.
`...`	Further parameters to be passed to `vennDiagram`.

Value

No return value, called for creating a plot.

Examples

	data(exampleData)
  # use the absolute correlation as a gene-level statistic
  stat <- abs(apply(countdata,1,cor,y = labels))
  # define the core set as the 25% genes with the highest correlation
  coreSet <- rownames(countdata)[tail(order(stat), 25)]
  
  # perform an overrepresentation analysis
  resOverrep <- geneSetAnalysis(
  	dat = countdata,
  	geneSets = pathways,
  	analysis = analysis.customOverrepresentation(),
  	coreSet = coreSet,
  	adjustmentMethod = "fdr")
  
  # plot a Venn diagram
  plotOverrepresentation(resOverrep, subset = 1:3, aggregate = TRUE)
data(exampleData)
  # use the absolute correlation as a gene-level statistic
  stat <- abs(apply(countdata,1,cor,y = labels))
  # define the core set as the 25% genes with the highest correlation
  coreSet <- rownames(countdata)[tail(order(stat), 25)]
  
  # perform an overrepresentation analysis
  resOverrep <- geneSetAnalysis(
  	dat = countdata,
  	geneSets = pathways,
  	analysis = analysis.customOverrepresentation(),
  	coreSet = coreSet,
  	adjustmentMethod = "fdr")
  
  # plot a Venn diagram
  plotOverrepresentation(resOverrep, subset = 1:3, aggregate = TRUE)

Predefined enrichment analyses

Description

Predefined analysis configurations that can be used in geneSetAnalysis

Usage

analysis.gsea()
analysis.overrepresentation()
analysis.customOverrepresentation()
analysis.averageCorrelation()
analysis.averageTStatistic()
analysis.globalTest()
analysis.globalAncova()
analysis.gsea()
analysis.overrepresentation()
analysis.customOverrepresentation()
analysis.averageCorrelation()
analysis.averageTStatistic()
analysis.globalTest()
analysis.globalAncova()

Details

The above functions return configurations for state-of-the-art analysis pipelines that can be used in geneSetAnalysis. All configurations are preconfigured collections of standard methods for the different pipeline steps. The following lists the methods chosen for the different steps and their parameters. For more detailed descriptions of these methods, please refer to the linked manual pages.

analysis.gsea defines the Gene Set Enrichment Analysis (GSEA) method by Subramanain et al. Here, the gene-level statistic the absolute correlation calculated by gls.cor with the associated parameters labs, method and a preprocessing by transformation.abs. As a gene set statistic, the enrichment score (function gss.enrichmentScore with parameter p) is calculated. The significance is assessed in a permutation test using significance.permutation with testAlternative = "greater" and free parameter numSamples, labs.
analysis.overrepresentation calculates an overrepresentation analysis using the gene-level statistic gls.tStatistic with parameters pValue(should be TRUE), alternative and labs. The resulting values are then transformed via transformation.adjustAndBinarize (parameters are the adjMethod and threshold). Finally gss.fisherExactTest is used as gene set statistic.
analysis.customOverrepresentation calculates an overrepresentation analysis using a user-defined core set coreSet. That is, instead of calculating this core set internally based on differential expression as the standard overrepresentation analysis, this function allows for defining custom core sets. It internally uses the global analysis global.overrepresentation.
analysis.averageCorrelation calculates the gene-level statistic as the absolute correlation using gls.cor (with parameters labs, method) and transformation.abs. The gene set statistic is the mean correlation calculated by gss.mean. The significance is assessed by comparing the gene set statistic to randomly sampled gene sets using significance.sampling (with the parameter numSamples and the preset parameter testAlternative = "greater").
analysis.averageTStatistic uses the absolute t statistic as the gene-level statistic by applying gls.tStatistic (with parameters labs, pValue, alternative) and transformation.abs. The gene set statistic is the mean t statistic in the gene set as returned by gss.mean. The significance is assessed by comparing the gene set statistic to randomly sampled gene sets using significance.sampling (with the parameter numSamples and the preset parameter testAlternative = "greater").
analysis.globalTest performs a global gene set enrichment analysis by Goeman et al. by applying the global.test function which in turn wraps the gt function in the globaltest package.
analysis.globalAncova applies the global ANCOVA method by Hummel et al. using the global method global.ancova which wraps the GlobalAncova function in the GlobalAncova package.

Value

All functions return an object of class gsAnalysis that specifies the corresponding analysis parameters for geneSetAnalysis.

References

Hummel, M., Meister, R., Mansmann, U. (2008) GlobalANCOVA: exploration and assessment of gene group effects. Bioinformatics, 24(1), 78–85.

Goeman, J. J., van de Geer, S. A., de Kort, F., van Houwelingen, H. C. (2004) A global test for groups of genes: testing association with a clinical outcome. Bioinformatics, 20(1), 93–99.

Examples


	data(exampleData)
  # apply a gene set analysis based on the average absolute correlation
  resAvCor <- geneSetAnalysis(
  	# parameters for geneSetAnalysis
  	dat = countdata,
  	geneSets = pathways[1],
  	analysis = analysis.averageCorrelation(),
  	adjustmentMethod = "fdr",
  	# additional parameters for analysis.averageCorrelation
    labs = labels,
  	method = "pearson",
  	numSamples = 10)
  
  # apply an overrepresentation analysis
  resOverrep <- geneSetAnalysis(
  	# parameters for geneSetAnalysis
  	dat = countdata,
  	geneSets = pathways,
  	analysis = analysis.overrepresentation(),
  	adjustmentMethod = "fdr",
  	# additional parameters for analysis.overrepresentation
  	pValue = TRUE,
  	threshold = 0.1,
  	labs = labels
  )
  
  # apply a global analysis using GlobalAncova
  resGA <- geneSetAnalysis(
  	# parameters for geneSetAnalysis
  	dat = countdata,
  	geneSets = pathways[1],
  	analysis = analysis.globalAncova(),
  	adjustmentMethod = "fdr",
  	# additional parameters for analysis.globalAncova
  	labs = labels,
  	method = "approx")

data(exampleData)
  # apply a gene set analysis based on the average absolute correlation
  resAvCor <- geneSetAnalysis(
  	# parameters for geneSetAnalysis
  	dat = countdata,
  	geneSets = pathways[1],
  	analysis = analysis.averageCorrelation(),
  	adjustmentMethod = "fdr",
  	# additional parameters for analysis.averageCorrelation
    labs = labels,
  	method = "pearson",
  	numSamples = 10)
  
  # apply an overrepresentation analysis
  resOverrep <- geneSetAnalysis(
  	# parameters for geneSetAnalysis
  	dat = countdata,
  	geneSets = pathways,
  	analysis = analysis.overrepresentation(),
  	adjustmentMethod = "fdr",
  	# additional parameters for analysis.overrepresentation
  	pValue = TRUE,
  	threshold = 0.1,
  	labs = labels
  )
  
  # apply a global analysis using GlobalAncova
  resGA <- geneSetAnalysis(
  	# parameters for geneSetAnalysis
  	dat = countdata,
  	geneSets = pathways[1],
  	analysis = analysis.globalAncova(),
  	adjustmentMethod = "fdr",
  	# additional parameters for analysis.globalAncova
  	labs = labels,
  	method = "approx")

Eliminate unknown genes from gene sets

Description

This function removes all genes that are not part of the experiment (not in rownames(dat)) from the specified gene sets which. All names are set to lower case.

Usage

preprocessGs(
	dat,
	geneSets)
preprocessGs(
	dat,
	geneSets)

Arguments

`dat`	A numeric matrix of gene expression values for all analyzed genes. Here, each row corresponds to one gene, and each column corresponds to one sample. The rows must be named with the gene names used in the gene sets. Here, only the row names (i.e. the gene names) are used by the function.
`geneSets`	A list of gene sets to be processed, where each gene set is a vector of gene names corresponding to the row names of `dat`.

Value

A list of preprocessed gene sets, where each gene set only contains those genes that are also present in dat

Examples

#values are not important, only the row names are used
dat <- matrix(0,100,10)
rownames(dat) <- paste("gene",1:100,sep="")

geneSets <- list(
		gs1 = paste("GENE",1:20,sep=""),# all genes in the analyzed data
		gs2 = paste("Gene",101:110,sep=""),#no gene in the analyzed data
		gs3 = paste("gene",90:110,sep="")#some genes in the analyzed data
	)

newGeneSets <- preprocessGs(dat = dat, geneSets = geneSets)
#values are not important, only the row names are used
dat <- matrix(0,100,10)
rownames(dat) <- paste("gene",1:100,sep="")

geneSets <- list(
		gs1 = paste("GENE",1:20,sep=""),# all genes in the analyzed data
		gs2 = paste("Gene",101:110,sep=""),#no gene in the analyzed data
		gs3 = paste("gene",90:110,sep="")#some genes in the analyzed data
	)

newGeneSets <- preprocessGs(dat = dat, geneSets = geneSets)

Significance assessment

Description

Functions to assess the significance of the gene-level statistics, as used in the significance parameter of gsAnalysis. These functions are based on applying the same analysis to randomly modified data sets or gene sets and comparing their statistic values to the original gene set statistic value.

Usage

significance.sampling(
	...,
	dat,
	geneSet,
	analysis,
	glsValues,
	numSamples = 1000)

significance.permutation(
	...,
	dat,
	geneSet,
	analysis,
	glsValues,
	numSamples = 1000,
	labs)

significance.restandardization(
	...,
	dat,
	geneSet,
	analysis,
	glsValues,
	numSamples = 1000,
	labs)
significance.sampling(
	...,
	dat,
	geneSet,
	analysis,
	glsValues,
	numSamples = 1000)

significance.permutation(
	...,
	dat,
	geneSet,
	analysis,
	glsValues,
	numSamples = 1000,
	labs)

significance.restandardization(
	...,
	dat,
	geneSet,
	analysis,
	glsValues,
	numSamples = 1000,
	labs)

Arguments

`...`	Additional parameters for the different steps of the analysis pipeline, depending on the concrete configuration supplied in `analysis`.
`dat`	The original data set as a numeric matrix of gene expression values for all analyzed genes. Here, each row corresponds to one gene, and each column corresponds to one sample. The rows must be named with the gene names used in the gene sets.
`geneSet`	The original gene set in form of a vector of gene names corresponding to the row names of `dat`.
`analysis`	The analysis applied to the original gene set (that should also be applied to the modified gene sets). This is an object of type `gsAnalysis` as produced by the function `gsAnalysis`.
`glsValues`	A vector containing the (possibly transformed) gene-level statistic values for each gene in the original data set `dat`.
`numSamples`	The number of random samples that should be taked to calculate the null distribution for the significance assessment. Default is `1000` for each test.
`labs`	A vector of class labels for the samples in `dat` for `significance.permutation` and `significance.restandardization`.

Details

Standard methods for the significance assessment of a gene set statistic (to be used in an analysis pipeline defined by gsAnalysis):

significance.sampling: This function repeatedly draws random gene sets. Their gene set statistic values form the null distribution.
significance.permutation: This function repeatedly permutes the labels of the data set. The gene set statistic values for the original gene set on the permuted data set form the null distribution.
significance.restandardization: This function applies both a gene set sampling and a label permutation. The permutation statistic values are standardized by their mean and standard deviation and then restandardized based on the gene set sampling statistic values. These restandardized values form the null distribution (Efron and Tibshirani).

Value

significance.sampling returns a list with the following elements:

`gssValues`	A vector of gene set statistic values, one entry per sample.
`randomGeneSets`	A matrix containing the gene sets which were sampled randomly from the set of all genes.

significance.permutation returns a list with the following elements:

`gssValues`	A vector of gene set statistics, one entry per sample.
`permutations`	A matrix, where each column contains the indices of one permutation.

significance.restandardization returns a list with the following elements:

`gssValues`	A vector of gene set statistics, one entry per sample.
`samplingValues`	A list of sub-lists, each containing one sampling result as defined above.
`permutationValues`	A list of sub-lists, each containing one permutation result as defined above.

References

Efron, B., Tibshirani, R. (2007) On testing the significance of sets of genes. Annals of Applied Statistics, 1, 107-129.

Summarize gene set analysis results

Description

Prints a summary of a gene set analysis result object.

Usage

## S3 method for class 'gsaResult'
summary(object,
		mode = c("summary", "table"),
		orderBy = c("adjustedPValues", "rawPValues", "geneSetName"),
		significantOnly = FALSE,
		signLevel = object$signLevel,
		...)
## S3 method for class 'gsaResult'
summary(object,
		mode = c("summary", "table"),
		orderBy = c("adjustedPValues", "rawPValues", "geneSetName"),
		significantOnly = FALSE,
		signLevel = object$signLevel,
		...)

Arguments

`object`	A result object as returned by `geneSetAnalysis`.
`mode`	Specifies the type of information that is displayed: By default (`mode="summary"`), a brief summary of the number of significant and insignificant gene sets is printed. For `mode="table"`, `createSummaryTable` is called, and a detailed table of adjusted and unadjusted p-values and the number of genes for each gene set is printed.
`orderBy`	If `mode="table"`, this specifies which field should be used for the row ordering. By default, rows are ordered according to the adjusted p-values.
`significantOnly`	If `mode="table"`, this specifies whether all gene sets (`significantOnly=FALSE`) or only the statistically significant gene sets (`significantOnly=TRUE`) should be included in the table.
`signLevel`	If `mode="table"` and `significantOnly=TRUE`, this specifies the significance level for the results that should be included in the table. By default, the original significance level of the analysis is used.
`...`	Currently unused

Value

Returns a data.frame with a summary of the produced results such as data set name, p-values, size of the given gene set.

Examples

data(exampleData)
  # perform gene set analyses for several pathways
  res <- geneSetAnalysis(
  	# global parameters
  	dat = countdata,
  	geneSets = pathways, 
  	analysis = analysis.averageCorrelation(), 
  	# additional parameters for analysis.averageCorrelation
  	labs = labels,
  	numSamples = 10)
  
  #summarize the analyses
  summary(res, mode = "summary")
  
  summary(res, mode = "table", orderBy = "rawPValues")
data(exampleData)
  # perform gene set analyses for several pathways
  res <- geneSetAnalysis(
  	# global parameters
  	dat = countdata,
  	geneSets = pathways, 
  	analysis = analysis.averageCorrelation(), 
  	# additional parameters for analysis.averageCorrelation
  	labs = labels,
  	numSamples = 10)
  
  #summarize the analyses
  summary(res, mode = "summary")
  
  summary(res, mode = "table", orderBy = "rawPValues")

Transformations

Description

Functions to transform the gene-level statistic values prior to the calculation of the gene set statistics, as used in the transformation parameter of gsAnalysis. Most of the functions wrap existing R functions.

Usage

transformation.abs(x)

transformation.square(x)

transformation.localFdr(x,
	statistic="pvalue",
	cutoff.method="fndr",
	pct0=0.75)

transformation.binarize(x, quant)

transformation.rank(x)

transformation.adjust(x, adjMethod = "fdr")

transformation.adjustAndBinarize(x, adjMethod = "fdr", threshold = 0.05)
transformation.abs(x)

transformation.square(x)

transformation.localFdr(x,
	statistic="pvalue",
	cutoff.method="fndr",
	pct0=0.75)

transformation.binarize(x, quant)

transformation.rank(x)

transformation.adjust(x, adjMethod = "fdr")

transformation.adjustAndBinarize(x, adjMethod = "fdr", threshold = 0.05)

Arguments

`x`	A numeric vector of gene-level statistic values, one per gene. These values are calculated by the previous step (see `gls`).
`statistic`	Specifies the null model for `transformation.localFdr` (see `statistic` parameter of `fdrtool` for possible values).
`cutoff.method`	Type of cut-off method used in `transformation.localFdr` (see `cutoff.method` parameter of `fdrtool` for possible values).
`pct0`	Fraction of data used by `transformation.localFdr` if `cutoff.method="pct0"` (see `fdrtool` for a detailed description).
`quant`	For `transformation.binarize`, this numeric value in the interval [0,1] defines the percentage of gene-level statistic values which should be set to zero. The remaining values are set to one.
`adjMethod`	The method to use for the adjustment for multiple testing (see `method` parameter of `p.adjust` for possible values).
`threshold`	The threshold for differential expression of a gene (defaults to `0.05`). Values smaller than these threshold are set to 1 , others to 0.

Details

Standard transformation functions for gene-level statistics (to be used in an analysis pipeline defined by gsAnalysis):

transformation.abs: Calculates the absolute values of the elements in x (a wrapper for abs).
transformation.square: Squares all elements in x.
transformation.localFdr: Calculates the local fdr for the elements in x. This is a wrapper for fdrtool.
transformation.binarize: Binarizes the values in x by using the quant quantile as a threshold.
transformation.rank: Ranks the values in x and returns the rank vector.
transformation.adjust: Adjusts for multiple testing according to the adjustment method specified in adjMethod.
transformation.adjustAndBinarize: Adjusts for multiple testing according to the adjustment method specified in adjMethod and binarizes the resulting p-values according to threshold (values smaller than the threshold become 1 others 0).

Value

All functions return a vector of transformed values having the same length as x.

Package 'GiANT'

Help Index

Enrichment analysis

Description

Details

Author(s)

References

Examples

Example data

Description

Usage

Details

Examples

Create an overview table for an analysis

Description

Usage

Arguments

Value

See Also

Examples

Quantify gene set uncertainty

Description

Usage

Arguments

Details

Value

See Also

Examples

Filtering of gene sets

Description

Usage

Arguments

Value

See Also

Examples

Main interface for enrichment analyses.

Description

Usage

Arguments

Details

Value

References

See Also

Examples

Gene-level statistics

Description

Usage

Arguments

Details

Value

See Also

Gene set statistics

Description

Usage

Arguments

Details

Value

References

See Also

Global analyses

Description

Usage

Arguments

Details

Value

References

See Also

Gene set analysis.

Description

Usage

Arguments

Details

Value

See Also

Examples

Null distribution histogram and statistic of the input set for enrichment analyses.

Description

Usage

Arguments

Details