R/annotatedPeak.R

#' Annotated Peaks
#' 
#' TSS annotated putative STAT1-binding regions that are identified in
#' un-stimulated cells using ChIP-seq technology (Robertson et al., 2007)
#' 
#' obtained by data(TSS.human.GRCh37)
#' 
#' data(myPeakList)
#' 
#' annotatePeakInBatch(myPeakList, AnnotationData = TSS.human.GRCh37,
#' output="b", multiple=F)
#' 
#' @name annotatedPeak
#' @docType data
#' @format GRanges with slot start holding the start position of the peak, slot
#' end holding the end position of the peak, slot names holding the id of the
#' peak, slot strand holding the strands and slot space holding the chromosome
#' location where the peak is located. In addition, the following variables are
#' included.  \describe{ \item{list("feature")}{id of the feature such as
#' ensembl gene ID} \item{list("insideFeature")}{upstream: peak resides
#' upstream of the feature; downstream: peak resides downstream of the feature;
#' inside: peak resides inside the feature; overlapStart: peak overlaps with
#' the start of the feature; overlapEnd: peak overlaps with the end of the
#' feature; includeFeature: peak include the feature entirely}
#' \item{list("distancetoFeature")}{distance to the nearest feature such as
#' transcription start site} \item{list("start_position")}{start position of
#' the feature such as gene} \item{list("end_position")}{end position of the
#' feature such as the gene} }
#' @keywords datasets
#' @examples
#' 
#' data(annotatedPeak)
#' head(annotatedPeak, 4)  # show first 4 ranges
#' if (interactive() || Sys.getenv("USER")=="jianhongou") {
#' y = annotatedPeak$distancetoFeature[!is.na(annotatedPeak$distancetoFeature)]
#' hist(as.numeric(as.character(y)), 
#'      xlab="Distance To Nearest TSS", main="", breaks=1000, 
#' ylim=c(0, 50), xlim=c(min(as.numeric(as.character(y)))-100, 
#' max(as.numeric(as.character(y)))+100))
#' }
#' 
"annotatedPeak"
LihuaJulieZhu/ChIPpeakAnno documentation built on Aug. 5, 2020, 12:02 a.m.