This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Getting differential ChIP-seq peaks between conditions after calling consensus peaks

Hello! I have some, pearhaps naive, questions about consensus peaks called using Tao Liu's script on GitHub called merge_then_call_consensus.sh.

Say, for example, I have 4 replicate samples and I manually find the overlapping peaks between them and keep the peaks with overlap in at least 3 of the 4 samples (using something like bedtools intersect). What will be the difference between these and consensus peaks called using the aforementioned script?

Also, say I have found consensus peaks across replicates for a TF ChIP in control versus stimulation conditions. I now want to get differential peaks between control and stimulation with the MAnorm tool. The tool requires that I provide the peak files as well as the read files. Can I use the consensus peak file of each condition with the read file produced by merging all the initial replicate bam files?

Or does it make more sense to merge the bam files for each condition, call peaks and then get differential peaks between control and stimulation?

Thanks in advance!

manorm chip-seq consensus peaks
• 2,100 views • link
written by Alexandra • Alexandra

1 answer

This has formally been investiated by a reputable group here: https://academic.oup.com/nar/article/42/11/e95/1442937

IIRC they think that merging bams and calling peaks on that is "best", given the choices of calling individually + merge, or joint calling.

Tbh, in practice it depends. If I have multiple groups and I need per-group peaks for downstream analysis I often just merge the individual peaksets. Or I call peaks per sample and keep peaks for the consensus that are called in at least x% of samples. It really depends. Maybe that's statistically not pretty, but it always depends on the project.

In the end, I make a count matrix based on the consensus peaks with featureCounts and analyze with DESeq2/edgeR/limma. I don't know MAnorm, but I doubt it's better than these three given how robust and established they are.

• 0 views • link
written by ATpoint ATpoint

Really interesting answer. I had missed that paper and will digest it asap. I usually use DiffBind, but recently have been looking into potential alternatives.

• 0 views • link
written by Ian Ian

That having said, linked paper and its authors developed the csaw package over at Bioconductor that uses a window approach to avoid peak calling altogether, and use the windows for DE analysis.

• 0 views • link
written by ATpoint ATpoint

Thank you for your response! Interesting paper.

• 0 views • link
written by Alexandra • Alexandra

Log in to answer this question.

More posts like this