Faster code for combining granges in CallPeaks()

[danielcgingerich]

Hi, I have noticed that for large datasets, CallPeaks() in signac takes a long time when specifying a grouping variable and subsequently using combine.peaks = TRUE. Running the source code, the culprit is the "for" loop that uses the revmap to match appropriate idents to the new gene ranges object. In my case, this takes a long time.

I have written a workaround for this that is vectorized and faster than the for loop. What are your thoughts? This is my approach:

gr.combined <- c(exc.peaks, inh.peaks, astro.peaks, micro.peaks, oligo.peaks, opc.peaks)

gr <- reduce(gr.combined, with.revmap = T, min.gapwidth = 1)

ident.vec <- vector(mode = 'character', length = length(gr))
exc.overlap <- findOverlaps(query = gr, subject = exc.peaks)
inh.overlap <- findOverlaps(query = gr, subject = inh.peaks)
astro.overlap <- findOverlaps(query = gr, subject = astro.peaks)
micro.overlap <- findOverlaps(query = gr, subject = micro.peaks)
oligo.overlap <- findOverlaps(query = gr, subject = oligo.peaks)
opc.overlap <- findOverlaps(query = gr, subject = opc.peaks)

ident.vec[queryHits(exc.overlap)] <- paste(ident.vec[queryHits(exc.overlap)], 'Exc', sep = ',')
ident.vec[queryHits(inh.overlap)] <- paste(ident.vec[queryHits(inh.overlap)], 'Inh', sep = ',')
ident.vec[queryHits(astro.overlap)] <- paste(ident.vec[queryHits(astro.overlap)], 'Astro', sep = ',')
ident.vec[queryHits(micro.overlap)] <- paste(ident.vec[queryHits(micro.overlap)], 'Micro', sep = ',')
ident.vec[queryHits(oligo.overlap)] <- paste(ident.vec[queryHits(oligo.overlap)], 'Oligo', sep = ',')
ident.vec[queryHits(opc.overlap)] <- paste(ident.vec[queryHits(opc.overlap)], 'OPC', sep = ',')

#remove the first comma
ident.vec <- substr(ident.vec, 2, nchar(ident.vec))


gr$peak.called.in <- ident.vec

[danielcgingerich]

@timoast

[timoast]

The code shown here seems fine as a workaround if you're finding the method used in the CallPeaks() function is slow. I haven't had any issues with this step being slow, but I can profile it more thoroughly when I get time and there may be better ways to do it.