Introduction and problem

I have multiple (>2) GRanges objects. I want to find those ranges that are shared by x% or more of all GRanges.

Example data

I will provide some example data as dataframes, let's say we want to find those ranges that are shared by 66.7% (2/3) or more.

gr1 <- data.frame(seqnames = rep('chr1', 3),
                  start = c(1, 10, 20), 
                  end = c(3, 17, 30))


gr2 <- data.frame(seqnames = rep('chr1', 3),
                  start = c(2, 11, 31), 
                  end = c(3, 19, 35))


gr3 <- data.frame(seqnames = rep('chr1', 3),
                  start = c(2, 16, 37), 
                  end = c(3, 22, 40))

Output wanted

A Granges output. In the example the algorithm should find:

chr1 2 - 3 Reason: (2-3 is found in gr1, gr2 and gr3, 1 only found in gr1) chr1 11 - 22 Reason: (11-17 is found in gr1 and gr2, 10 only in gr1 ,18-19 in gr2 and gr3, 20 -22 in gr1 and gr3)

What I have done

I know how to find query hits found in all (100%) GRanges, see https://stackoverflow.com/questions/23331475/r-overlap-multiple-granges-with-findoverlaps

Is this a coverage question?

Here's an approach with plyranges, after converting these df to ranges with as_granges():

n <- 3 # number of range sets                                                                                                    
bind_ranges(gr1, gr2, gr3, .id="origin") %>%                                                                                                   
  compute_coverage() %>%                                                                                                         
  mutate(fraction_cov = score / n) %>%                                                                                           
  filter(fraction_cov > .66) %>%                                                                                                 
  reduce_ranges()

There are some details, but you can work from here. E.g. this assumes that the individual ranges in the incoming range sets don't overlap each other. You could call group_by(origin) and reduce_ranges on those from the outset, after binding them together.