comparisons between 2 ChIP-seq datasets

Question

[Bogdan]

Hi everyone,

We have two sets of ChIP-seq data for a histone mark (lets' say H3K4me1) that were generated with :

<> an antibody Ab1

<> an antibody Ab2

What is the best method to compare these two ChIP-seq datasets in order to be able to say :

Ab1 works better than Ab2 ?

Thanks,

Bogdan

Answer 1

[jared.andrews07]

Honestly, I'd just eyeball the tracks in IGV. It's usually pretty apparent.

If you want more objective measures, you could look at total peaks called, fraction of reads in peaks, compare peak overlap with public datasets in similar cells/tissue if available, etc.

[Bogdan]

Yes, there is an entire ENCODE metrics that we can use. The issue is that NSC or RSC do not discriminate so well between ChIP-seq samples on histone marks that have moderate level of noise.

Answer 2

[ATpoint]

You do not have any replication, hence you cannot make any meaningful statistics, nor can you make statements about the range of noise the antibodies create between experiments with the same AB batch or between LOTs. Hence, just look at the browser tracks and count peaks and FRiPs as suggested already. In the absence of IP replication one AB would need to be 'much better' than the other for a decision on which to follow up with and 'much better' would mean that one produces like 10000 quality peaks and the other only half. Nuance differences can be technical noise, I often see notable differences in peak numbers and FRiPs even with the same AB in the same experiment. ChIP just has so many sources of bias, and without replication there is no way to assess AB performance meaningfully. Taken together, take the AB that either works obviously better when it comes to peak numbers and FRiPs and by looking at the data, or if that is not clearly different then the one that is likely to have better longterm supply or simply the cheaper one, that's how I usually tackle this.

Cross-posted https://support.bioconductor.org/p/9144367/#9144367