Is ChIP-Seq quantitative?

[Livi81]

Hi all,

Can we make between library comparisons of peak height? FindPeaks was used to identify sites of transcription factor binding using the same antibody in two different cell types. If peak height is lower in one library than the other at a certain promoter can we infer that different levels of the transcription factor (or histone modification for that matter) are at that locus? would the libaries have to be sequenced to the same depth with the same mapping efficiencies to make any comparison? Does anyone know how reproducible peak heights are between replicate samples?

Any suggestions/discussions/questions welcome!

Thanks

[kenosaki]

You mean, comparing the peaks of two libraries without input control?

[Livi81]

Hi,
Sorry I should have mentioned that we've already subtracted input control from both libraries. I was just wondering if one library has a peak of say height 20 and another library has a peak at the same region but at height 80. Can we extrapolate any biological meaning from that? RNA-Seq is normalised to library depth and gene length to give quantative data. Are ChIP-Seq libraries normalised in any sort of similar ways?

Thanks again

[biznatch]

I think you need to consider the overall number of reads in the library which I guess is what you mean by library depth in RNA-seq. Eg. if one library has 20 million and the other has 10, then I'd expect that peaks in the 20 million library would be on average twice the height of the ones in the 10 million library. I don't know if that means you should just divide everything by 2 in the 20 million library. I'd also be interested in knowing what the best way is to normalise data for comparisons like this.

[HESmith]

In addition to normalizing for read depth, there's also the issue of how the peak number differs between samples. For example, if your transcription factor binds 1000 sites in sample A and 2000 in sample B then, all else being equal, the peak height would be twice as high in sample A for the same number of reads.

[Livi81]

You're right. I guess the conclusion is that using current methods ChIP-Seq is only qualitative. It's a shame I think we're missing out on what could be interesting information.

[wangxin]

So, what is the most important information we get when we do the ChIP-seq?

[Livi81]

That the protein of interest is interacting with the underlying DNA sequence in a given cell type or tissue.