[BioC] question about DiffBind

Rory Stark Rory.Stark at cruk.cam.ac.uk
Fri May 16 16:36:56 CEST 2014


Hi Rong-

You are correct that the peak merging logic results in wider peak areas
that may incorporate multiple binding sites and/or make it more difficult
to determine the precise binding site. This is true of the peak calling
step as well. In our experience, the scientific questions we are actually
asking do not require the precise binding location in the differential
analysis step -- we are first looking for places where there is a
significant difference in binding levels. It is only after that, and only
in a minority of cases, that we look to narrow the region down to a
specific binding site of interest (eg, to find enriched motifs). Tehre are
a numebr of tools available to find more narrow, precise binding sites in
a set of candidate regions.

The most recent release of DiffBind does, however, include some features
to better a) focus on the likely binding site and b) work with more
uniform, narrower peak widths. It accomplishes this using the "summits"
parameter in dba.count. This parameter allows you to identify the "summit"
for each peak interval (the point of highest pileup), and optionally
re-center the peaks around this point (eg, taking 200bp upstream and
downstream of the summit for each merged peak; summits=200). If there were
several "true" binding sites in a merged peak, you are likely to end up
with only one of them, but the results are more consistent overall.

Cheers-
Rory

On 12/05/2014 17:40, "Rong Cong" <rong.cong at inserm.fr> wrote:

>Hi Rory,
>Thank you very much for your patient and efficient reply to my naive
>question. The short version answer made me quite clear of the idea.
>
>Indeed, overlapping and merging peaks may form a quite wider peak to be
>compared among different peaksets, but for biologists, the wider peak
>often has no meaning to us(some biologically functional elements are
>quite short),therefore is it possible for DiffBind to adjust some
>parameters to narrow a little the overlapped and merged peak for
>comparison? (Again, I am not a bioinformatician, sorry for this naive
>question.)
>Looking forward to your reply.
>All the best,
>Rong
>



More information about the Bioconductor mailing list