How to facilitate the performance of overlap for GRanges objects (verifiable example is given)?

Hi every one:

I have GRanges objects to find overlap simultaneously, I did implement function to accomplish this task, it works good so far. Now I need to figure out better way to facilitate the process, and needs to come up more compatible function to improve the re-usability of my code. However, I tried my attempt to get better function for it, but I don't happy with my current function performance. Therefore, I came for this community to get some ideas to achieve my desired output.

Here is the reproducible example to understand the problem that I faced:

data:

gr1 <- GRanges(
      seqnames=Rle(c("chr1", "chr2", "chr3", "chr4"), c(5, 6, 4, 3)),
      ranges=IRanges(seq(1, by=9, len=18), seq(6, by=9, len=18)),
      rangeName=letters[seq(1:18)], score=sample(1:25, 18, replace = FALSE))

gr2<- GRanges(
  seqnames=Rle(c("chr1", "chr2", "chr3","chr4"), c(4, 7, 5, 4)),
  ranges=IRanges(seq(2, by=11, len=20), seq(8, by=11, len=20)),
  rangeName=letters[seq(1:20)], score=sample(1:25, 20, replace = FALSE))
gr3 <- GRanges(
  seqnames=Rle(c("chr1", "chr2", "chr3","chr4"), c(8, 7, 6, 4)),
  ranges=IRanges(seq(4, by=11, len=25), seq(9, by=11, len=25)),
  rangeName=letters[seq(1:25)], score=sample(1:25, 25, replace = FALSE))

step 1:

library(magrittr)
gr1.ss <- gr1 %>% subset(score > 12) 
gr1.ww <- gr1 %>% subset(score >5 & score <=12)
gr1.nn <- gr1 %>% subset(score <= 5)

 targets <- list(gr2, gr3)

step 2 :

ov.ss <- lapply(targets, function(ele_) {
  res <- as(findOverlaps(gr1.ss, ele_), "List")
})
ov.ww <- lapply(targets, function(ele_) {
  res <- as(findOverlaps(gr1.ww, ele_), "List")
})

this is my attempt to facilitate ov.ss, ov.ww can happen one wrapper function, so I can loop through gr1.res by finding match of its name, and internally call ov .

step 3 - my attempt (need to be improved) :

gr1.res <- list('ss'=gr1.ss, 'ww'=gr1.ww, 'noise'=gr1.nn)
for(i in 1:length((gr1.res))) {
  query <- S4Vectors::getListElement(gr1.res, i)
  if(names(query)=='nn') {
     message("do not process, and directly export as data.frame")
     out <- unique(query) %>% as.data.frame
  }
  ov <- lapply(targets, function(ele_) {
      res <- as(findOverlaps(query, ele_), "List")
    })
  ov
}

I don't want to process gr1.nn and just directly export them as data.frame objects. I want to integrate doing ov.ss and ov.ww in one wrapper functions, so I want to come up new way to do this. How can I facilitate my attempt solution? How can I have more compatible function for facilitating my desired attempt. I will be grateful if any one can give me any enlightening solution, possible suggestion or ideas to achieve my desired functions. Thanks a lot

Best regards:

Jurat

r granges overlap performance function

• 2,309 views • link

written 10.1 years ago by Jurat Shahidin

1 answer

What about the following:

> intersect(intersect(gr1, gr2), gr3)
GRanges object with 6 ranges and 0 metadata columns:
      seqnames    ranges strand
         <Rle> <IRanges>  <Rle>
  [1]     chr1  [ 4,  6]      *
  [2]     chr1  [15, 15]      *
  [3]     chr1  [19, 19]      *
  [4]     chr1  [28, 30]      *
  [5]     chr1  [37, 41]      *
  [6]     chr2  [92, 96]      *
  -------
  seqinfo: 4 sequences from an unspecified genome; no seqlengths

Just to make sure that the order of the intersections does not matter:

> intersect(intersect(gr3, gr2), gr1)
GRanges object with 6 ranges and 0 metadata columns:
      seqnames    ranges strand
         <Rle> <IRanges>  <Rle>
  [1]     chr1  [ 4,  6]      *
  [2]     chr1  [15, 15]      *
  [3]     chr1  [19, 19]      *
  [4]     chr1  [28, 30]      *
  [5]     chr1  [37, 41]      *
  [6]     chr2  [92, 96]      *
  -------
  seqinfo: 4 sequences from an unspecified genome; no seqlengths

• 1 views • link

written 10.1 years ago by Giovanni M Dall'Olio

Dear Giovanni M Dall'Olio:

Thanks again your quick respond on my post. I am pretty sure about step 2 and it is very fast and way efficient. Regarding your solution, it is very important for me to preserve its metadata like p.value, global fisher score and so on, and I did implement function for those.

The point is I need to do multiple testing on my data to observe test variation in different test in the context of Biological replicates, and each test would yield different output as well. The matter is, each test, I need to classify query GRanges (a.k.a, gr1 in my data) into three subgroup by its score (key idea is to remove background noise before doing overlap), then apply step 2 that I showed above, order is really matters also.

However, I am seeking more compatible way to iterate through gr1.ss, gr1.ww, gr1.nn (gr1.nn refers to noise) in the list, and find its overlap respectively, so I come up step 3, but it don't happy with it. Could you give me some interesting ideas to make this procedure better? Thanks a lot.

Best regards:

Jurat

• 0 views • link

written 10.1 years ago by Jurat Shahidin