Hi,
I have ChIP-Seq data where my Input sample is sequenced to 15M reads and IP sample to 45M reads. Is read count normalisation a good way to normalise the IP with Input? If not, whats the best way to proceed with these Input and IP samples of different sequencing depths?
Thanks a lot!
Before normalizing the IP to the input (input subtraction I guess), I recommend using the signal extraction scaling (SES) method, which has been implemented in deeptools bamCompare. It separates reads into signal and background before read depth normalization (on the background reads only), improving the signal-to-noise ratio.
Compare (a - SES normalization) with (b- total read normalization):
Does that ensure comparability between IPs, e.g. what TMM from edgeR is doing, so correcting for compositional differences between libraries? Problen with using TMM here is that it is so different from the IP that the general assumptions fail, I usually ignore inputs for that reason once peaks are called.
Probably not, IMHO the SES method is more like a "within sample" normalization (IP - input of the same sample). From what I read of the OP, he has only one IP sample and its matching input sample, and he wants to correct from potential chromatin bias (spurious peaks in the input) before visualization.
Apologies, if I was not clear. I do have multiple IP samples, each with a corresponding Input sample. Just the Input samples were sequenced at a third of depth than the IP samples. I would like to look at enrichment of my protein of interest in IP samples at certain genomic loci by plotting heatmaps.
Even so the SES method above should work well for visualization and in particular if you want to look at each IP-input heatmap separately. On the other hand, if you want to statistically compare the IP between them, you should look at tools that do differential peak calling, but this is beyond the scope of your question.
If you are ok with omitting the inputs then you can normalize the IPs using ATAC-seq sample normalization to get scaled bigwigs and then use these for plotting. That is my preferred approach to make sure the compositional and signal/noise differences between IPs are properly corrected for.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
What do you want to do, call peaks?
I would like to plot heatmaps at certain genomic loci, highly transcirbed genes.
Do you want to plot heatmaps for both IP and input OR plot a heatmap of the IP normalized to the input ?
Hi, its just to plot a heatmap of the IP normalized to the Input.