Question: DiffBind TMM normalization
0
3.4 years ago by
akdess0
akdess0 wrote:

Hi,

I am using DiffBind to assess the H3K27me3 chipseq binding differences between two conditions. I am using default settings. I would like to visualize differentially bound regions in IGV/UCSC Genome Browser. I have generated RPM normalized bigwig files for visualization. However some of the sites that DiffBind identified, don't look differentially bound when I look at those regions in IGV. The reason is probably because DiffBind does TMM normalization, which is different than RPM normalization. My question is that how can I plot the signal and show differentially bound regions? Is there a way to generate TMM normalized bigWig files from bam files? My other question is that DiffBind gives me around 2000 regions that are differentially bound and almost all of them have positive fold change in one condition, is it normal? When I look at the mean RPKM fold changes they are not correlated with dba.report output fold changes. Is it expected?

Thanks,

Akdes

modified 3.3 years ago by Rory Stark2.8k • written 3.4 years ago by akdess0
0
3.3 years ago by
Rory Stark2.8k
CRUK, Cambridge, UK
Rory Stark2.8k wrote:

Hi Akdes-

Sorry for the delay in responding!

I can't think of an easy way to get TMM normalized bigWig files, as the normalization occurs over read counts aggregated over the entire peak intervals.

It may help to look at the dba.report()s using bNormalized=FALSE to see the raw read counts. Also, compare  dba.plotMA() with bNormalized=TRUE and with bNormalized=FALSE to see the impact of the normalization. If your experiment really does involve lots of changes all in one direction, then the TMM normalization method form edgeR will not really be appropriate. You may want to re-try the analysis using the DESeq2 normalization, which is more conservative in these conditions (which is the reason that DESeq2 has become the default method for DiffBind analysis in the current release).

Cheers-

Rory