Hi!
I have peaks called by MACS on 4 biological replicates.
What I want, is to know the number of differential regions in these replicates.
What are the tools available or what approach can be applied to get these regions?
Other than DESeq and DiffBind what else can be used?
Thank you
diffReps can analyze your ChIP-seq data in an automated pipeline. In addition to finding differential sites, it can also identify hotspots and annotate your differential sites/peaks to genes or heterochromatic regions. diffReps does not depend on peak calling, though. This actually provides you with better results because peak calling is not very well defined for histone modifications, or any protein that generates broad peaks. What's more, differences can often be found within histone modification peaks. If you use peak calling, you will fix the regions that you want to compare, and then lose the potentially interesting regions. diffReps uses a sliding window strategy so you won't miss a dime.
We had favorable experience with diffReps. It can even identify differential sites when those peak calling programs failed. When the peaks do not stand out from background clearly, a peak calling program may have difficulty in identifying them. However, diffReps uses the information from multiple biological replicates to strengthen the stat. When we applied MACS to a very difficult histone mark in neurons, it can only identify a small number of peaks and the pathway analysis does not make much sense. However, diffReps was able to identify a decent number of changes between our treatment and control and the pathway analysis indicated that they may be real targets.
diffreps is the one I have used in addition to DiffBind. I think they are both pretty good.
Edit: Some others from an earlier answer I gave at SeqAnswers
DBChIP, which appears to use edgeR internally
GLM method in the Kasowski et al Science paper
Apparently, CisGenome can do this as well
Very interesting topic. I am trying MAnorm.
MAnorm only considers the intensity or confidence of a peak but not its shape. Depending on the variation of the replicates, the peaks may have different sizes: some are longer, some are shorter and some are shifting towards up- or down-stream. This is OK if you just want to get a consensus peak list under basal condition. But if you want to compare two different conditions, how can you define a "consensus" to ensure fair comparison?
As asked, it seems you want to identify the regions that are called differently by MACS between your replicates. One method of identifying these regions would be to use bedtools. Since MACS reports peaks as a bed file you could simply subtract bed files from each other in successive steps to identify regions unique to each replicate, e.g.
bedtools subtract -a rep1_peaks.bed -b rep2_peaks.bed -A
Another approach, although you say you don't want to use DESeq, would be to use it to investigate the regions called by the replicates in your experiment. Since MACS also outputs a "summits.bed" file containing the base location of the summit of each peak, It would be pretty easy to read those bed files into R, add some flanking distance to each summit to create segments (i.e. plus or minus 100 bases), and then use DESeq to generate p-values for every segment. Since you would know which segments were contributed by which replicates, and which are shared by various replicates, you'd be able to explore several characteristics of your peaks and how the replicate behave.
There is a downside with this kind of approach. Given the variation between different replicates, it can become very difficult to define the so called "consensus" peak. You may come up with various strategies but no one can tell which gives you the best result. In addition, +/-100bp probably works well with DNA binding transcription factors but definitely NOT with histone marks.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
Hello, Do you also merge the differential binding regions among these different tools?
Thanks!