GO term enrichment analysis for large numbers of gene sets

To test a sample population of genes for over-representation of GO terms, the core function GOHyperGAll computes for all nodes in the three GO networks (BP, CC and MF) an enrichment test based on the hypergeometric distribution and returns the corresponding raw and Bonferroni corrected p-values. Subsequently, a filter function supports GO Slim analyses using default or custom GO Slim categories. Several convenience functions are provided to process large numbers of gene sets (e.g. clusters from partitioning results) and to visualize the results.

Note: GOHyperGAll provides similar utilities as the GOHyperG function in the GOstats package. The main difference is that GOHyperGAll simplifies processing of large numbers of gene sets, as well as the usage of custom array-to-gene and gene-to-GO mappings.

## Generate gene-to-GO mappings and store as catDB object 
makeCATdb(myfile, lib = NULL, org = "", colno = c(1, 2, 3), idconv = NULL, 
            rootUK=FALSE)

## Enrichment function
GOHyperGAll(catdb, gocat = "MF", sample, Nannot = 2)

## GO slim analysis
GOHyperGAll_Subset(catdb, GOHyperGAll_result, sample = test_sample, 
            type = "goSlim", myslimv)

## Reduce GO term redundancy 
GOHyperGAll_Simplify(GOHyperGAll_result, gocat = "MF", cutoff = 0.001, correct = TRUE)

## Batch analysis of many gene sets
GOCluster_Report(catdb, setlist, id_type = "affy", method = "all", CLSZ = 10, 
            cutoff = 0.001, gocats = c("MF", "BP", "CC"), myslimv = "default",
            correct = TRUE, recordSpecGO = NULL, ...)

## Bar plot of GOCluster_Report results
goBarplot(GOBatchResult, gocat)

Arguments

myfile: File with gene-to-GO mappings. Sample files can be downloaded from geneontology.org (http://geneontology.org/GO.downloads.annotations.shtml) or from BioMart as shown in example below.
colno: Column numbers referencing in myfile the three target columns containing GOID, GeneID and GOCAT, in that order.
org: Optional argument. Currently, the only valid option is org="Arabidopsis" to get rid of transcript duplications in this particular annotation.
lib: If the gene-to-GO mappings are obtained from a *.db package from Bioconductor then the package name can be specified under the lib argument of the sampleDFgene2GO function.
idconv: Optional id conversion data.frame
catdb: catdb object storing mappings of genes to annotation categories. For details, see ?"SYSargs-class".
rootUK: If the argument rootUK is set to TRUE then the root nodes are treated as terminal nodes to account for the new unknown terms.
sample: character vector containing the test set of gene identifiers
Nannot: Defines the minimum number of direct annotations per GO node from the sample set to determine the number of tested hypotheses for the p-value adjustment.
gocat: Specifies the GO type, can be assigned one of the following character values: "MF", "BP" and "CC".
GOHyperGAll_result: data.frame generated by GOHyperGAll
type: The function GOHyperGAll_Subset subsets the GOHyperGAll results by directly assigned GO nodes or custom goSlim categories. The argument type can be assigned the values goSlim or assigned.
myslimv: optional argument to provide custom goSlim vector
cutoff: p-value cutoff for GO terms to show in result data.frame
correct: If TRUE the function will favor the selection of terminal (informationich) GO terms that have at the same time a large number of sample matches.
setlist: list of character vectors containing gene IDs (or array feature IDs). The names of the list components correspond to the set labels, e.g. DEG comparisons or cluster IDs.
id_type: specifies type of IDs in input, can be assigned gene or affy
method: Specifies analysis type. Current options are all for GOHyperGAll, slim for GOHyperGAll_Subset or simplify for GOHyperGAll_Simplify.
CLSZ: minimum gene set (cluster) size to consider. Gene sets below this cutoff will be ignored.
gocats: Specifies GO type, can be assigned the values "MF", "BP" and "CC".
recordSpecGO: argument to report in the result data.frame specific GO IDs for any of the 3 ontologies disregarding whether they meet the specified p-value cutoff, e.g: recordSpecGO=c("GO:0003674", "GO:0008150", "GO:0005575")
GOBatchResult: data.frame generated by GOCluster_Report
...: additional arguments to pass on

Details

GOHyperGAll_Simplify: The result data frame from GOHyperGAll will often contain several connected GO terms with significant scores which can complicate the interpretation of large sample sets. To reduce this redundancy, the function GOHyperGAll_Simplify subsets the data frame by a user specified p-value cutoff and removes from it all GO nodes with overlapping children sets (OFFSPRING), while the best scoring nodes are retained in the result data.frame.

GOCluster_Report: performs the three types of GO term enrichment analyses in batch mode: GOHyperGAll, GOHyperGAll_Subset or GOHyperGAll_Simplify. It processes many gene sets (e.g. gene expression clusters) and returns the results conveniently organized in a single result data frame.

Value

makeCATdb generates catDB object from file.

References

This workflow has been published in Plant Physiol (2008) 147, 41-57.

Author

Thomas Girke