IP efficiency in Chip-Seq data

Question

[dnamonk]

Hi,

I am new to this forum. I am curious if there is a way (possibly a program) to fix IP efficiency issue across chip replicates?

Thanks in advance.

[ivivek_ngs]

you want to fix the IP efficiency or you want to test if the IP quality based on signal to noise ration. You can use SPP cross-correlation analysis and CHANCE for the same. This will give you all the information about the quality of the IP.

This will only let you know the quality and then your peak calling, in that case, might be lenient and not done on stringent thresholds.

Answer 1

[Devon Ryan]

You can't really "fix" the differences, the best you can do is to try to include it when normalizing vs. input, for example with the SES normalization in bamCompare from deepTools.

Answer 2

[Michele Busby]

By "fix" do you mean making the less efficient experiment more efficient?

The less efficient experiment will have a higher signal to noise ratio, meaning that there will be more background noise. You cannot increase the signal to noise ratio above what is there because you cannot efficiently remove the noise.

I would ask:

Why is the signal to noise ratio high in the low efficiency experiment? If this is your lab's data there is a lot you can do to optimize the protocol including adjusting the input amounts, etc.

Was the same antibody used in all the reps? If the efficiency is different because it is not the same antibody (or the same lot of the a polyclonal antibody) there may be bigger differences than the efficiency. We did some work on this here: https://epigeneticsandchromatin.biomedcentral.com/articles/10.1186/s13072-016-0100-6 if you want to see what I mean.

Did one of the immunoprecipitations just fail? There are degrees of bad and if it outright failed you may get more signal by throwing out the failed rep.

So if you have some clue about why some are worse than others it is easier to think about what to do next. The protocols for ChIP Seq are so fiddly.

[Devon Ryan]

The less efficient experiment will have a lower signal to noise ratio, not higher :)

[Michele Busby]

yeah, what Devon said.

Answer 3

[dnamonk]

Found this paper for the benefit of other people! enRich R package: https://www.rdocumentation.org/packages/enRich/versions/3.0 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3717085/

They try to address the IP efficiency issue!

Thank you all for your suggestions

Cheers !

[Devon Ryan]

They don't claim to fix anything, they just add efficiency to their model, which is only applicable in multi-sample peak calling.

[dnamonk]

I agree with you. They just add efficiency to their model.