% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/pca2go.R
\name{topGOtable}
\alias{topGOtable}
\title{Extract functional terms enriched in the DE genes, based on topGO}
\usage{
topGOtable(
  DEgenes,
  BGgenes,
  ontology = "BP",
  annot = annFUN.org,
  mapping = "org.Mm.eg.db",
  geneID = "symbol",
  topTablerows = 200,
  fullNamesInRows = TRUE,
  addGeneToTerms = TRUE,
  plotGraph = FALSE,
  plotNodes = 10,
  writeOutput = FALSE,
  outputFile = "",
  topGO_method2 = "elim",
  do_padj = FALSE
)
}
\arguments{
\item{DEgenes}{A vector of (differentially expressed) genes}

\item{BGgenes}{A vector of background genes, e.g. all (expressed) genes in the assays}

\item{ontology}{Which Gene Ontology domain to analyze: \code{BP} (Biological Process), \code{MF} (Molecular Function), or \code{CC} (Cellular Component)}

\item{annot}{Which function to use for annotating genes to GO terms. Defaults to \code{annFUN.org}}

\item{mapping}{Which \code{org.XX.eg.db} to use for annotation - select according to the species}

\item{geneID}{Which format the genes are provided. Defaults to \code{symbol}, could also be
\code{entrez} or \code{ENSEMBL}}

\item{topTablerows}{How many rows to report before any filtering}

\item{fullNamesInRows}{Logical, whether to display or not the full names for the GO terms}

\item{addGeneToTerms}{Logical, whether to add a column with all genes annotated to each GO term}

\item{plotGraph}{Logical, if TRUE additionally plots a graph on the identified GO terms}

\item{plotNodes}{Number of nodes to plot}

\item{writeOutput}{Logical, if TRUE additionally writes out the result to a file}

\item{outputFile}{Name of the file the result should be written into}

\item{topGO_method2}{Character, specifying which of the methods implemented by \code{topGO} should be used, in addition to the \code{classic} algorithm. Defaults to \code{elim}}

\item{do_padj}{Logical, whether to perform the adjustment on the p-values from the specific
topGO method, based on the FDR correction. Defaults to FALSE, since the assumption of
independent hypotheses is somewhat violated by the intrinsic DAG-structure of the Gene
Ontology Terms}
}
\value{
A table containing the computed GO Terms and related enrichment scores
}
\description{
A wrapper for extracting functional GO terms enriched in the DE genes, based on
the algorithm and the implementation in the topGO package
}
\details{
Allowed values assumed by the \code{topGO_method2} parameter are one of the
following: \code{elim}, \code{weight}, \code{weight01}, \code{lea},
\code{parentchild}. For more details on this, please refer to the original
documentation of the \code{topGO} package itself
}
\examples{
library("airway")
library("DESeq2")
data("airway", package = "airway")
airway
dds_airway <- DESeqDataSet(airway, design= ~ cell + dex)
# Example, performing extraction of enriched functional categories in
# detected significantly expressed genes
\dontrun{
dds_airway <- DESeq(dds_airway)
res_airway <- results(dds_airway)
library("AnnotationDbi")
library("org.Hs.eg.db")
res_airway$symbol <- mapIds(org.Hs.eg.db,
                            keys = row.names(res_airway),
                            column = "SYMBOL",
                            keytype = "ENSEMBL",
                            multiVals = "first")
res_airway$entrez <- mapIds(org.Hs.eg.db,
                            keys = row.names(res_airway),
                            column = "ENTREZID",
                            keytype = "ENSEMBL",
                            multiVals = "first")
resOrdered <- as.data.frame(res_airway[order(res_airway$padj),])
de_df <- resOrdered[resOrdered$padj < .05 & !is.na(resOrdered$padj),]
de_symbols <- de_df$symbol
bg_ids <- rownames(dds_airway)[rowSums(counts(dds_airway)) > 0]
bg_symbols <- mapIds(org.Hs.eg.db,
                     keys = bg_ids,
                     column = "SYMBOL",
                     keytype = "ENSEMBL",
                     multiVals = "first")
library(topGO)
topgoDE_airway <- topGOtable(de_symbols, bg_symbols,
                             ontology = "BP",
                             mapping = "org.Hs.eg.db",
                             geneID = "symbol")
}

}