I would like to concatenate individual genomic intervals into common regions. 

My input: 

   dfin <- "chr start end sample type
            1   10    20   NE1    loss
            1   5     15   NE2    gain
            1   25    30   NE1    gain
            2   40    50   NE1    loss
            2   40    60   NE2    loss
            3   20    30   NE1    gain"
    dfin <- read.table(text=dfin, header=T)

My expected output: 

    dfout <- "chr start end samples type
            1   5     20   NE1-NE2  both
            1   25    30   NE1      gain
            2   40    60   NE1-NE2  loss
            3   20    30   NE1      gain"
    dfout <- read.table(text=dfout, header=T)

The intervals in dfin will never overlap in the same animal, just between animals (columns sample and samples, respectively). The column type have two factors (loss and gain) in dfin and is expected to have three factors in dfout (loss, gain and both, which occur when the concatenated region in dfout was based on both loss and gain).   

Any idea to deal with that?

library(GenomicRanges)
gr <- as(dfin, "GRanges")

out_gr <- reduce(gr, with.revmap=TRUE)
out_samples <- extractList(as.character(mcols(gr)$sample),
                           mcols(out_gr)$revmap)
out_samples <- as.factor(unstrsplit(sort(out_samples), sep="-"))
out_type <- extractList(as.character(mcols(gr)$type),
                        mcols(out_gr)$revmap)
out_type[elementLengths(out_type) == 2] <- "both"
out_type <- factor(as.character(out_type),
                   levels=c(levels(mcols(gr)$type), "both"))
mcols(out_gr) <- DataFrame(samples=out_samples, type=out_type)
dfout <- as.data.frame(out_gr)
dfout
#   seqnames start end width strand samples type
# 1        1     5  20    16      * NE1-NE2 both
# 2        1    25  30     6      *     NE1 gain
# 3        2    40  60    21      * NE1-NE2 both
# 4        3    20  30    11      *     NE1 gain

Cheers,

H.