Hi,
I ran diffbind and used edgeR and DESeq2.
I have a question regarding the normalized counts. In the DESeq2 analysis, the normalized counts of one of the samples always had integers. In EdgeR - there was no column with integers.
In EdgeR normalization - one column is chosen as reference to all, and in DESeq2 they create another column of the averages that is ussed for normalization.
So In EdgeR I would expect to have after normalization one column with integers, but not in DESeq. I actually get the opposite.
Can you help with this please?

Also, in the publication of the diffbind EdgeR was the default. Now DESeq2 is the default.
Do you have any preference to one of the methods?

Thanks a lot.
 

To answer the second question first, the default analysis method was changed from edgeR to DESeq2 in the current release due to its more conservative normalization. The default TMM normalization in edgeR may over-normalize when a large number of features all change in the same direction, which is more likely in a ChIP-seq than an RNA-seq experiment. When affinity changes occur in both directions, the TMM normalization works well.

Regarding the one-column-of-integers issue, I assume you are looking at the count scores reported from dba.report() with bScores=TRUE? Those values are calculated from within DiffBind using normalization values computed by edgeR and/or DESeq2, as follows.

For edgeR values, the read counts (which may have control reads subtracted, and set to a minimum value of 1) are divided by a normalisation factors derived by multiplying the lib.size by the norm.factors. Only in certain cases will this yield a column of integers.

For DESeq2 values, the reads (adjusted for control reads and minimum value as above) are divided by the result of calling sizeFactors(). 

Regards-

Rory

Thanks a lot for your quick reply and for the great package.

Regarding the normalized counts - I am referring to the counts that are obtained after calling dba.report() with bCounts=TRUE.
I ran DiffBind several times and always for DESeq2 one of the columns has integers.
When using DESeq2 normalization for RNA Seq I never get a column of integers (and I wouldn't expect to get integers, since each count is divided by the size factor).
So how is it I always get a column of integers for DESeq2 in DiffBind?

Thanks

Thank you very much.
So if I don't expect lots of changes between the samples, it might be better to set bFullLibrarySize=FALSE right?
 

Yes, if there isn't a big imbalance in the signal between sample groups, bFullLibrarySize=FALSE is preferred. This used to be the default, but we changed it to be more conservative as the consequences are worse that way if the assumptions are not met.

-R

Thanks a lot!