Hi Rory Stark,

I am a little confused about the following questions regarding the use of DiffBind:

How does DiffBind handle normalization? Does it use the raw counts from BAM files? Are the total reads counts used for normalization those in peak regions? Also, when we choose DeSeq2 medthod, does it normalize the data as we analyze RNA-seq read counts, which require raw read counts as input?

Is it possible to use DiffBind without spike-in normalization?

Thanks in advance.

Yes, you can do normalization without spike-in reads, using the raw counts from the BAM files.

The default normalization method in DiffBind is the minimal, "first-do-no-harm" method of computing normalization factors based solely on the ratios of the total number raw reads in the supplied BAM files.

You can also use reads large background bins by calling dba.normalize() and setting background=TRUE.

You can chose to normalize based only on the reads overlapping peak regions by setting library=DBA_LIBSIZE_PEAKREADS (or library="RiP"), for Reads-in-Peaks). In general this is not advisable unless you are confident that most of the consensus peaks do not have big changes in binding affinity.

There is a detailed chapter in the DiffBind vignette that examines the different normalization options in detail. Also, the help page for dba.normalize() may be useful.