batch effect for ChIP-seq data

Question

[Ming Tommy Tang]

Hi everyone,

If I want to compare ChIP-seq data from different sequencing projects, say epigenome roadmap vs ENCODE.

How do you normalize across samples? Is it similar to RNA-seq data that one needs to correct batch effect http://simplystatistics.org/2015/05/20/is-it-species-or-is-it-batch-they-are-confounded-so-we-cant-know/

I know MAnorm http://bcb.dfci.harvard.edu/~gcyuan/MAnorm/MAnorm.htm and others can do for samples from same project.

I just want to know how you deal with it for ChIP-seq data.

Thanks,
Ming

Answer 1

[jotan]

I generally do not normalise for ChIP-seq although I would be curious to know if others do. The ChIP protocol itself is highly variable and different modifications/transcription factors can also have vastly different binding profiles, so I'm not certain that applying normalisation would be meaningful. I usually call peaks independently and overlap, only normalising for total read counts.

[Ming Tommy Tang]

How about comparing the same histone mark among different samples from different sequencing centers.

[jotan]

That's pretty tricky. I would only be comfortable doing that for very robust histone modifications like H3K4me3. Otherwise, the variability will be quite high. There are also a few other points to keep in mind.

1. ChIP-seq is not really quantitative. I don't think anyone knows what the signal strength actually represents.
2. The ChIP protocols are not uniform and most labs often have their own signature protocol.

For a robust histone modification, I would normalise for total read counts (randomly extract reads to match the smallest file), call peaks individually and overlap the peaks.

[Ming Tommy Tang]

Thanks for sharing your tips!