Hello everyone! I have a question about a histone marks ChIP-seq experiment. When I normalized my samples with RPKM or TMM, I realized WT condition has a higher signal everywhere (background and peaks detected by macs2). So although the signal-to-noise ratio is lower than the mutant condition, the signal in the peaks is higher, even in regions I had checked by RT-qPCR before sequencing. I was wondering If there is a way to normalize the samples according to the background of each one, or any normalization method to solve this problem.
Normalize for what purpose? If you have a background for each sample, for visualization it's not uncommon to convert coverage depth of each to RPM and take a moderated log2 ratio of ip/input. What is it that you want to do with the values?
Since nucleosomes bind the chromatin across most of the genome, it breaks a lot of assumptions made when using traditional ChIPseq normalization methods like NCIS (Liang, 2012) that were built for transcription factor(TF) targets that more sparsely bind the genome.
You could use annotations to quantify the amount of background signal. Try finding a set of nucleosome free region(NFR) annotations in your organism and sum the coverage in small windows (the goal is to capture as much of the NFR region without running into the "signal" peaks) around those regions to come up with a scaling factor for each sample that makes sure the NFRs scale to more/less the same magnitude.
The major assumption to keep in mind with this is that the NFR annotations are the same for both WT and mutant samples. If nucleosome positional shifts are moving nucleosomes into the NFR regions, this breaks the normalization assumption. If you can filter the NFR annotations to sites known to have very fixed nucleosome positioning, that can mitigate some of that concern. Perhaps you can diagnose this by performing ChIPseq peak calling in each of the samples, matching up the corresponding peaks in each sample and check the distribution of these distances. The distribution should be tight around zero if the mutant does not affect positional changes to the nucleosome.
Also, remember that comparing magnitudes of binding between samples is always tricky and you should keep to comparing between samples that use the same target/antibody. Even more so if you use different methods of normalization so if you compare these samples against TF-target ChIPseq data, you can really only compare the positional information (i.e. peak positions).
Cross-posted to Bioconductor https://support.bioconductor.org/p/9142702/
Normalize for what purpose? If you have a background for each sample, for visualization it's not uncommon to convert coverage depth of each to RPM and take a moderated log2 ratio of ip/input. What is it that you want to do with the values?