Identify cell types based-on RNA-seq/ChIP-seq data

Question

[dcheng1]

Hi! I was wondering if there are any methods that can identify cell-type-specific ChIP-seq signal?

I've tried edgeR, and some packages based on edgeR, such as diffbind or csaw. However, those methods only do pairwise comparison. But I want to compare more than 3 groups altogether. For example, I have ChIP-seq/RNA-seq data from 3 different cell types, I want to know the location where signal predominantly shows up in one type of cell, but not in the other two? Then I can use those cell-type-specific ChIP-seq signals to deconvolute heterogeneous tissue samples.

I think I can still do pairwise comparison, such as find the differentially binding(DB) sites between A vs. B, and A vs. C, and take the intersection between the two DB subsets as DB specific to cell type A. Does this sounds solid? Or do you know any direct way to do this? Thanks a lot!!

Answer 1

[Devon Ryan]

In practice most people would use a loose p-value cut off and perform intersections as you described. There are problems with that and you're probably better served using an appropriate contrast in CSAW and using only high-fold change results (you'd compare the cell type of interest to the average of the other groups). That would likely fail for very large numbers of cell types (due to an inflated false discovery rate), but otherwise it should work well enough in practice.

[dcheng1]

I will try that. Thanks!!