Strandedness Of Homer For Motif Analysis

Question

[user]

I'm using homer on a bunch of regions that have been called as having peaks to do motif discovery. Homer (http://biowhat.ucsd.edu/homer/ngs/peakMotifs.html) docs say that it needs a strand parameter. Since the regions are from a ngs experiment where the library sequenced is unstranded, the strand has no meaning and my peaks are unstranded. In that case, what's the correct way to feed the regions to Homer? If I give it a BED with no strand value in the 4th column, it appears to run but I have no idea how it handles this case. Will the -rna parameter make it ignore strand? Ultimately I want it to avoid strandedness when determining the motif.

I think that MEME can be made to ignore strand but not sure about Homer.

[keryruo]

Have you solved this problem? Can you give some advice?

[keryruo]

I've tried setting 3 different arguments for the strand(+-.),after comparison of the HOMMER annotationannotatePeaks.pl peak.bed hg19 > peak.anno ),results related to these 3 strand arguments were all the same.So you guys don't have to worry about the strand infor

Answer 1

[Istvan Albert]

The sequenced DNA fragment will produce reads from the 5' end of each strand separately. So even though the library was not strand specific the reads themselves can actually be separated by strand and the peaks can and should be called by strand.

The reason that the strand information is important is that these strand specific peaks will actually determine the length of the bound location, without them one is left with mere guessing.

There are methods that merge reads to produce unstranded peaks, in that case you can still split them by strand and create artificial boundary peaks at the expected distances, of course the accuracy of the method above will of course suffer greatly.

[user]

They can only be called by strand if they are mapped to known stranded regions unambiguously, like genes.

[Istvan Albert]

but they are mapped as such - I think you mean that the directionality of the promoter/bound region is unknown. That is true. One would not immediately know which way the transcription goes. But the information on the two strands that actually formed the fragments is unambiguous because the sequencing proceeds from the 5' end. The left border will be formed by hits to the positive strand and the right border will be formed by reads mapping to the reverse strand.

[user]

Yes, agreed, but your method relies on using the borders of the peaks to tell the strand. I meant that for a single read in isolation, the strand can't be known, unless it maps let's say in the promoter region of a gene that's in the + strand in which case it can be inferred

Answer 2

[Ming Tommy Tang]

I had the same question? anyone can help?

my peaks are called by MACS and no strand information was given.

[Istvan Albert]

please don't ask a new question when you are posting an answer. MACS does not know which strand your motif binds to. It simply shows you where the region is, you have to determine the orientation from other sources of information.