[BioC] diffbind question : bSubControl with DEseq
Rory.Stark at cruk.cam.ac.uk
Thu Jan 9 12:39:38 CET 2014
With the default setting of bScaleControl=TRUE in the call to dba.count,
the number of control reads will be scaled down prior to subtraction if
there are more control reads than ChIP reads. However the reverse is not
also true: no scaling occurs if there are fewer reads in the control
library, so the impact of subtracting the control reads is less than it
might be. In general, the SubControl feature doesn't do much except where
there are very large spikes in the control, so it shouldn't make too much
difference; the bottom line is that we didn't want to scale down the
primary ChIP signal without a good reason.
As the newCountDataset object is generated by DESeq, it doesn't make sense
to use RPKM here. The DESeq authors are very clear that only true,
non-normalised read counts should be used. You can work with RPKM values
for the plotting features of DiffBind (by setting score=DBA_SCORE_PRKM in
the plotting function, or in dba.count, where the score can be reset at
will if peak=NULL). You can also retrieve the RPKM values for any sample,
or for the entire binding matrix, using dba.peakset with bRetrieve=TRUE.
On 08/01/2014 18:30, "aurélie teissandier" <aurelie.teissandier at curie.fr>
>I use diffbind with Deseq. And I look in the newCountDataSet object ,
>the counts for each peak.
>By default, the option bSubControl is true. So for one peak in one
>condition (condition C1), the count is egal to ReadsCount in C1 -
>ReadsCount in Control .
>Is is true? Before the subtraction, there are not normalization between
>condition C1 and control samples?
>So for example, if you sequence twice more reads in C1 compared to
>control, may be you have to divide the read count in C1 by 2.
>Can I use RPKM in the newCountDataset object? How?
>Thanks in advance,
>26, rue d'Ulm - 75248 Paris Cedex 05 - FRANCE
>Tel: 01 56 24 69 26
More information about the Bioconductor