Union and intersection of intervals

问题

I have a group of intervals for different ids. For example:

df <- data.frame(id=c(rep("a",4),rep("b",2),rep("c",3)), start=c(100,250,400,600,150,610,275,600,700), end=c(200,300,550,650,275,640,325,675,725))

The intervals of each id do not overlap but the intervals of the different ids may overlap. Here is a picture:

plot(range(df[,c(2,3)]),c(1,nrow(df)),type="n",xlab="",ylab="",yaxt="n")
for ( ii in 1:nrow(df) ) lines(c(df[ii,2],df[ii,3]),rep(nrow(df)-ii+1,2),col=as.numeric(df$id[ii]),lwd=2)
legend("bottomleft",lwd=2,col=seq_along(levels(df$id)),legend=levels(df$id))

What I'm looking for is for two functions: 1. A function which will take the union of these intervals. For the example above, it will return this data.frame:

union.df <- data.frame(id=rep("a,b,c",4), start=c(100,400,600,700), end=c(325,550,675,725))

A function which will intersect these intervals, only keeping a range if all the ids overlap for that range. For the example above, it will return this data.frame:

intersection.df <- data.frame(id="a,b,c", start=610, end=640)

回答1:

This is a bit awkward, but the idea is that you unroll the data into a series of opening and closing events. Then you track how many intervals are open at a time. This assume each group doesn't have any overlapping intervals.

df <- data.frame(id=c(rep("a",4),rep("b",2),rep("c",3)), start=c(100,250,400,600,150,610,275,600,700), end=c(200,300,550,650,275,640,325,675,725))


sets<-function(start, end, group, overlap=length(unique(group))) {
    dd<-rbind(data.frame(pos=start, event=1), data.frame(pos=end, event=-1))
    dd<-aggregate(event~pos, dd, sum)
    dd<-dd[order(dd$pos),]
    dd$open <- cumsum(dd$event)
    r<-rle(dd$open>=overlap)
    ex<-cumsum(r$lengths-1 + rep(1, length(r$lengths))) 
    sx<-ex-r$lengths+1
    cbind(dd$pos[sx[r$values]],dd$pos[ex[r$values]+1])

} 

#union
with(df, sets(start, end, id,1))
#     [,1] [,2]
# [1,]  100  325
# [2,]  400  550
# [3,]  600  675
# [4,]  700  725

#overlap
with(df, sets(start, end, id,3))
#      [,1] [,2]
# [1,]  610  640

回答2:

The intervals package solves the union part of the question:

require(intervals)
idf <- Intervals(df[,2:3])
as.data.frame(interval_union(idf))

And for the intersect part, depending on how the intervals are defined:

idl <- lapply(unique(df$id),function(x){var <- as(Intervals(df[df$id==x,2:3]),"Intervals_full");closed(var)[,1]<- FALSE;return(var)})
idt <- idl[[1]]
for(i in idl)idt <- interval_intersection(idt,i)
res <- as.data.frame(idt) 
res
   V1  V2
1 610 640

回答3:

For the intersection, I would start by counting the number of intervals you're in at each range (the beginning of the range is labeled with ord.dirs$x in this code and the number of intervals in the range is ord.dirs$z):

dirs <- data.frame(x=c(df$start, df$end), y=rep(c(1, -1), each=nrow(df)))
ord.dirs <- dirs[order(dirs$x),]
ord.dirs$z <- cumsum(ord.dirs$y)
ord.dirs <- ord.dirs[!duplicated(ord.dirs$x, fromLast=T),]
ord.dirs
#      x  y z
# 1  100  1 1
# 5  150  1 2
# 10 200 -1 1
# 2  250  1 2
# 14 275 -1 2
# 11 300 -1 1
# 16 325 -1 0
# 3  400  1 1
# 12 550 -1 0
# 8  600  1 2
# 6  610  1 3
# 15 640 -1 2
# 13 650 -1 1
# 17 675 -1 0
# 9  700  1 1
# 18 725 -1 0

Now you just need to grab the ranges where you have the correct number of intervals (3 in this case):

pos.all <- which(ord.dirs$z == length(unique(df$id)))
data.frame(start=ord.dirs$x[pos.all], end=ord.dirs$x[pos.all+1])
#   start end
# 1   610 640

You can similarly use ord.dirs to grab the union of the sets:

zero.pos <- which(ord.dirs$z == 0)
data.frame(start=c(ord.dirs$x[1], ord.dirs$x[head(zero.pos, -1)+1]),
           end=ord.dirs$x[zero.pos])
#   start end
# 1   100 325
# 2   400 550
# 3   600 675
# 4   700 725

回答4:

The GenomicRanges package provide some intersect and overlap funtions:

library(GenomicRanges)
source("http://bioconductor.org/biocLite.R")
biocLite("Gviz")    
library(Gviz)

make a Grange object with equal seqnames (this is important)

df <- data.frame(id=c(rep("a",4),rep("b",2),rep("c",3)),     start=c(100,250,400,600,150,610,275,600,700), end=c(200,300,550,650,275,640,325,675,725))
gr <- GRanges(seqnames = rep(1,nrow(df)),IRanges(start = df$start,end =      df$end))

Now you can plot the ranges with the Gviz package, as well.

d0 <- GenomeAxisTrack()
d1 <- AnnotationTrack(gr,group = df$id,fill=df$id)
plotTracks(c(d0,d1))

The union is done via reduce where intervals are collapsed

as.data.frame(reduce(gr))[,2:3]

the intersect is done via findoverlaps. Afterwards, filterd by ranges which overlaps 3 ranges.

OL <- as.data.frame(findOverlaps(gr,type="within"))
table(OL[,1])

df[as.numeric(names(which(table(OL[,1])==3))),]

来源：https://stackoverflow.com/questions/30079720/union-and-intersection-of-intervals

标签

intervals