Hello everyone,
I would like to do a differential binding analysis on a cut&run experiment I performed with "roughly" 2 replicates using 23 antibodies on Treatment vs DMSO ctrl with ecoli spike in normalization but unsure how to implement the normalization.
I say "roughly" because a sample failed library prep in replicate 1 and I had to repeat it along with a few histones in a 3rd replicate (to make sure the biology is behaving similarly).
I used ecoli spike-in dna in my experiments but obviously, I cannot assume the same amount of spike in-between replicates.
There seems to be 2 approaches A and B for DB analysis (from what I understood reading the forum):
A) pool bams together (all treatment bams across replicates together and same for DMSO bams together but separate from treatment)-->
get peaks -->
merge across replicates defining consensus peakset -->
count reads in said peaks -->
perform DESeq2/edgeR differential count analysis.
I am not sure how to proceed regarding incorporating spike in % normalization after pooling the bams for A):
I am following the custom macs2 peak calling guide and can scale the bedgraph files using the bedtools genomecov --scale option but Example.: If ecoli for H3K27ac was at 4% for DMSO replicate 1 (with 6M uniquely mapping reads) and 20% for replicate 2 (with 9M uniquely mapping reads) what is the scaling factor to input into the bedtools genomecov --scale command if I pool both into 1 bam?
B) csaw for an unbiased window based strategy. a section is available in the csaw book but any caveats or recommendations would be appreciated regarding use of spikein normalization (I vaguely remember reading some members having reservations for the use of spike ins in this stated use case).
Finally,
I am not sure if my treatment causes major global chromatin changes as the RNAseq effects are not as pronounced and the treatment does not affect the cells in-vitro/vivo in a major way (w.r.t proliferation/apoptosis).
But it does modulate some lineage markers so take that assumption with a grain of salt.
I only mention this because the csaw paper states the choice of normalization depends on this expectation (similar/same? to DESeq2 that we don't expect most genes to be DE).
Thank you for this forum resource.
I learnt a lot just by reading other posts.
Best,
Karim
Honestly, I would just use default edgeR/DESeq2 normalization and then look at the MA-plots for the individual comparisons. That will tell you whether it is normalized well. The majority of points should be somewhat y=0. I do not see the advantage of spike-ins.
Look at the data in the IGV. That is always the first I do, get an idea whether it even produced distinct meaningful peaks and how overall the changes look.