Use of IDR after running MACS3 for ATAC-seq data

Question

[Orange]

Cross-posted from github (a little worried about the inactivity there; please let me know if this is not good practice)

I am analyzing ATAC-seq data from human cells. I am planning to perform the following as part of my pipeline:

1. Adapter and T-overhang trimming with Cutadapt
2. Mapping to reference genome with bowtie2
3. PCR duplicate removal (Picard), quality and blacklist filtering (sambamba or samtools)
4. Peak calling with MACS3

Looking at online pipelines, the use of IDR post peak calling with MACS2 seems to be recommended. Is this still the case with the recently released MACS3 v3.0.0 when used with either of the following modes?

• default: macs3 callpeak -f BAMPE -t ATAC.bam -g hs -n test -B -q 0.01, or
• hmmratac: macs3 hmmratac

Also, when IDR is used, MACS2 seems to be run with p-value cutoff of 0.1. If the use of IDR is still recommended post MACS3, should it be run with -p 0.1 as well?

Thanks in advance for your help!

Answer 1

[ATpoint]

Is this still the case with the recently released MACS3 v3.0.0 when used with either of the following modes?

I use it downstream of "macs3" which mainly offers speed improvements afaik. The peak calling algorithm did not change afaik substantially.

MACS2 seems to be run with p-value cutoff of 0.1. If the use of IDR is still recommended post MACS3, should it be run with -p 0.1 as well?

Yes, one should use a relaxed pvalue (not q-value) cutoff since IDR (based on my superficial understanding) needs a good portion of both "good" and "bad" peaks to make an accurate call on which of these are reproducible between datasets, hence you should not only input peaks like FDR < 10^-5 because then the "noise" aspect that the model needs to learn is gone.

Back in the day I learned with this respource: https://hbctraining.github.io/Intro-to-ChIPseq/lessons/07_handling-replicates-idr.html HBC has a lot of great resources.

[Orange]

Thanks for your suggestion, I will check out the linked resource! (I am still learning what the best analysis pipeline might be and I most likely have more questions but I will make them into a separate post...)