Adapt metadata of a GenomicRanges object when disjoined

Question

I have a GRanges object with some genomic intervals and some metadata (3 vectors with the coverage of each region in 3 different samples). I have applied:

disjoin(my_data)

to obtain a new GRanges object with the smallest set of unique, non-overlapping pieces.

The problem is that I cannot conserve metadata in my new GRanges object. What I would like to obtain is the mean coverage of genomic regions which included this unique set.

As an example, I would like to turn this metadata:

       sample1   sample2   sample3
1:1-3    30        NA         NA
1:1-4    NA        40         35
1:4-5    35        NA         NA
1:5-7    NA        50         50
1:6-7    60        NA         NA 

into this:

       sample1    sample2     sample3
1:1      30         40          35
1:2      30         40          35
1:3      30         40          35
1:4      35         40          35
1:5      35         50          50
1:6      60         50          50
1:7      60         50          50

How can I achieve that?

Answer 1

Here is a data.table approach to conserving metadata for the disjoined set of ranges.

library(GRanges)
library(data.table)
data.disjoin <- disjoin(my_data)
overlaps <- as.data.frame(findOverlaps(data.disjoin,data))
coverage.disjoin <- as.data.table(cbind(overlaps,mcols(my_data)[overlaps$subjectHits,]))
coverage.disjoin <- coverage.disjoin[,
                      lapply(.SD[,-1],function(x){unique(x[!is.na(x)])}),
                      by="queryHits"]
mcols(data.disjoin) <- coverage.disjoin[,.(sample1,sample2,sample3)]

First, find the overlaps between the disjoined set of ranges and the original data. Then collect the coverage for the overlaps into a data.table. Find the unique coverage for that range by sample, removing NA values. Note that .SD is a special symbol for the subsetted data.table for the group. Finally, join the result back onto the disjoined data.

Data

my_data <- GRanges(
  c("chr1","chr1","chr1","chr1","chr1")
  ,IRanges(c(1,1,4,5,6),c(3,4,5,7,7)),
  sample1=c(30,NA,35,NA,60),
  sample2=c(NA,40,NA,50,NA),
  sample3=c(NA,35,NA,50,NA))