I have used the TMM method in CSAW package to normalize the composition bias for my different ChIP-seq samples and also got the normalized bigwig files. Next step I want to analyze the differential binding peaks between several mutant with WT for specific IP (for example, H3K4me1). But I was confused about how to choose the proper way to do the differential binding peaks analysis. What I know now, there exist two basic strategies I could use:
option 1. I use the DiffBind package to do that. The initial input file is bam files and a series of bed files generated by MACS2 call peaks (such as narrowPeak format). In this package, I will generate the normalization factors for the raw bam files before starting the differential binding analysis. option 2. I could also use the strategy in the CSAW package to filter out the low abundance region and only keep the higher abundance region for multiple tests between different samples. (as shown in the second example mentioned in this paper(https://f1000research.com/articles/4-1080/v1). In this example, I do not need to do peak calling beforehand and I just by calculating the counts by setting the bin width as a narrow or wide range.
Based on these, I have several questions:
- since I want to use TMM composition bias normalization to normalize all bam files for specific IP, if I choose the DiffBind package to get the differential binding peaks based on raw bam files, will it cause some argument since in both TMM composition bias normalization in CSAW and DiffBind will generate their own normalization factors.
- is the peak calling by MACS2 is an essential step before the differential binding peak analysis? can I omit the peak calling step, just use the generate matrix generated by TMM composition bias normalization to do the differential binding peak analysis?
- if choose the DiffBind to analyze differential binding peaks, for the bed files of two biological replicates, should I use the common regions or the pooling regions of two biological replicate to do the analysis?
- If I pool the coordinate of peaks of all samples (macs2 callpeak) and calculate the count matrix based on the TMM composition bias-normalized bigwig files, is it possible to import the matrix directly into DiffBind to do the differential binding peak analysis?
I would appreciate it if someone would like to give me some suggestions on that! Many thanks.