R/findFeatureComb.R

In BIMSBbioinfo/genomation: Summary, annotation and visualization of genomic data

# ---------------------------------------------------------------------------- #
#' Find combitations of genomic features
#' 
#' Provided a GRangesList, finds the combinations of sets of ranges. 
#' It is mostly used to look at the combinatorics of transcription factor 
#' binding. The function works by, firstly, constructing a union of all 
#' ranges in the list, which are then designated by the combinatorics of 
#' overlap with the original sets. A caveat of this approach is that the 
#' number of possible combinations increases exponentially, so we would 
#' advise you to use it with up to 6 data sets. If you wish to take a 
#' look at a greater number of factors, methods like self organizing 
#' maps or ChromHMM might be more appropriate.
#'
#' @param gl a \code{GRangesList} object, containing ranges for which 
#'                 represent regions enriched for transcription factor binding
#' @param width \code{integer} is the requested width of each enriched region. If 0 the ranges are not resized, if a positive integer, the width of all ranges is set to that number. Ranges are resized relative to the center of original ranges.
#' @param use.names a boolean which tells the function whether to return the resulting ranges with a numeric vector which designates each class (the default), or to construct the names of each class using the names from the GRangesList
#' @param collapse.char a character which will be used to separate the class names if use.names=TRUE. The default is ':'
#'
#' @usage findFeatureComb(gl, width=0, use.names=FALSE, collapse.char=':')
#' @return a GRanges object
#'
#'@examples
#' library(GenomicRanges)
#' g = GRanges(paste('chr',rep(1:2, each=3), sep=''), IRanges(rep(c(1,5,9), times=2), width=3))
#' gl = GRangesList(g1=g, g2=g[2:5], g3=g[3:4])
#' findFeatureComb(gl)
#' findFeatureComb(gl, use.names=TRUE)
#'
#' @docType methods
#' @rdname findFeatureComb-methods
#' @export
setGeneric("findFeatureComb", 
           function(gl, width=0, use.names=FALSE, collapse.char=':') 
             standardGeneric("findFeatureComb") )

#' @aliases findFeatureComb,GRangesList-method
#' @rdname findFeatureComb-methods
setMethod("findFeatureComb", signature("GRangesList"),
          function(gl, width, use.names, collapse.char){
          
              if(width < 0)
                stop('width needs to be a positive integer')
              
              # resizes the ranges before doing the overlaps
              if(width > 0)
                gl = endoapply(gl, function(x)resize(x, width=width, fix='center'))
                
              # makes a union of ranges and calculates the combinatorics
              r = reduce(unlist(gl))
              do = data.frame(lapply(gl, function(x)
                                            as.numeric(countOverlaps(r, x) > 0)))
              # finds the combinations of ranges for each interval in the reduced ranges
              do.comb = as.numeric(factor(rowSums(t(t(do)*(2^(0:(ncol(do)-1)))))))
              
              if(use.names == FALSE){
                values(r)$class = do.comb
              
              }
              if(use.names == TRUE){
                
                cnames = colnames(do)
                # selects only unique representatives of combinations
                do = do[!duplicated(do.comb),]
                # pastes names of individual sets
                cnames = apply(do, 1, function(x)paste(cnames[x == 1], 
                                                  collapse=collapse.char))
                
                # designates each reduced set by the combination of overlapping ranges
                r$class = cnames[match(do.comb, unique(do.comb))]
                
              }
              return(r)
          }
)