Calling differential peaks on large set of 3 conditions

Question

[14134125465346445]

Hi all,

What are the recommended methods to call differential peaks for enrichment/pulldown datasets (e.g. MEDIPS, H3K4me3) of different conditions with many replicates?

• 20-30 biological replicates of condition1 pulldown+input pairs
• 20-30 biological replicates of condition2 pulldown+input pairs
• 20-30 biological replicates of condition3 pulldown+input pairs

So far I have seen the following:

• macs2 bdgdiff: for pairs of condition1 vs condition2 peaks called with macs2 callpeak https://github.com/taoliu/MACS/wiki/Call-differential-binding-events

This gives me the right output, e.g. regions enriched in condition1 vs condition2 and vice versa, as well as common.

The question is how do I combine my 20-30 biological replicates of condition1 and my 20-30 biological replicates of condition2 in macs2 bdg files to run the bdgdiff pairwise comparison mode?

• THOR-2: seems to be able to do more than 2 in #rep1 and #rep2 lists http://www.regulatory-genomics.org/thor-2/tool-usage/

• Other tools to investigate:

https://omictools.com/polyphemus-tool https://omictools.com/differential-identification-using-mixtures-ensemble-tool https://omictools.com/mmdiff-tool https://omictools.com/genome-wide-generalized-additive-model-tool https://omictools.com/normr-obeys-regime-mixture-rules-tool https://omictools.com/chromstar-tool

• Other tools I discarded for this purpose:

  • Segway, ChromHMM, ChromImpute: don't seem to have a differential mode.
  • Epicode: differential mode seems to give differential "codes" as output, instead of regions (peaks).
  • PARE: seems to call differential regions with no nucleosomes, rather than differential peaks.

Thanks

Answer 1

[Fluorine]

I would recommend DiffBind R package which in my hands works best for MEDIPS or ChIP-seq analysis with replicates and/or multiple conditions. MACS2 bdgdiff also works well, but then I would do an additional step afterward to see which are the common peaks between replicates, which you can do with IDR analysis. You can combine your samples by combining alignment SAM or BAM files with samtools merge.

THOR should, in theory, do the same as DiffBind but I don't have much experience with it.

[Friederike]

DiffBind is basically a wrapper around DESeq2 with parameters optimally tuned for ATAC-seq data, i.e. based on how you set up the contrast it will use the replicates to gauge the variability per condition.

csaw is a similar wrapper around edgeR, which also offers the same type of syntax for specifying arbitrarily complex comparisons.

[Fluorine]

FYI DiffBind uses both DESeq and EdgeR, based on user's choice.