How is chip seq quantitative?

Question

[Maya]

Hello everyone I have a little issue of understanding and I was hoping someone can explain this to me in a simple way. I have seen some papers where they have used chip seq peak height to predict gene expression and some other papers where they look for histone QTLs (where they find SNP which associate with the height of peaks). I think this is really cool but I am confused how this is possible and what the "height" of the chip seq peak actually means. When I do chip seq isn't it just saying that in this one cell there is a histone modification right here and that's it? It's not like you can have strong or weak histone modifications. I can understand if you do chip seq in a tissue which is composed of lots of cells maybe there is cases where the histone modification is present in many cells which might increase the height of the peak. Is this correct? If that's the case does that mean we can't do this type of analysis in single cell data?

Thanks in advance

Answer 1

[Rory Stark]

One way to think about this in bulk ChIP-seq is that we're measuring the proportion of cells that have the protein bound (or histone mark) at each location. This is not exact, especially if there are copy number issues, but if each cell has a set number of copies of a given sequence of DNA, then seeing more reads associated with this region indicates that more of the chromatin in the sample was bound.

Peak-calling -- identifying genomics regions where there is evidence of protein binding (or chromatin marks etc.) -- is generally not quantitative; it is more of a binary yes/no (with quantitative confidence statistics). To look for binding changes (other than the difference between no binding and "some" binding), you can look for systematic differences in the read densities in specific genomic intervals. So if a factor was bound in 25% of the chromatin in one sample group (enough to be identified as a peak), and it was bound in 90% of a second sample group, we can calculate the confidence that this change in binding affinity is real quantitatively, provided we have enough replicates to capture the technical and biological variance.