ChIP-seq normalization for input with RPKM

Question

[gewa]

Hi all, I'm currently trying to work off part of a paper (found here) involving processing some ChIP-seq data for histone modifications (ie H3K4me1). At one point (under "Histone Modication Inputs, Normalization, Preprocessing") the authors say:

The ChIP-seq reads of these histone modifications were binned into 100bp intervals and normalized against its corresponding inputs by using an RPKM (reads per kilobase per million) measure.

I'm a bit confused as to how this could be replicated and what it means. My initial thought was to use a tool like deeptools bamCoverage to get a normalized (RPKM) value for 100bp bins. However, my understanding is that this is a normalization for sequencing depth - not a normalization against the ChIP-seq input. I thought I should then use bamCompare to compare the input vs targeted ChIP-seq, but it occurs to me that this will give me a ratio, not a RPKM value per bin. Could anyone explain what they mean by "normalized against it's corresponding inputs" in this case? Sorry if any of this is naive, I'm new to this and appreciate any help.

Answer 1

[ATpoint]

The idea is that you bin ChIP and Input into the same bins/intervals. Then you normalize the reads overlapping the bins by RPKM for both ChIP and Input. Doing so, you account for differences in sequencing depth (not discussing now if RPKM is a good choice, because there are multiple posts and papers out on this). After normalizing, you could use the log2-ratio between the bins as a measurement of enrichment over input. Better and more elaborate approaches are e.g. the BioC package csaw, which also relies on a binning/sliding window strategy, while applying elaborate statistics and control of false-discovery rate. It also has a outstandingly comprehensive manual that is worth reading.