[ChIP-seq. signal plot] How to align the based line of different samples?

Question

[chiefcat]

Hi all,

Does anyone have an idea how to make the baseline of different samples to the same level in a ChIP-seq. signal plot?

I've been using deepTools for normalization and plotting (have tried different normalization method provided in deepTools e.g. 1x normalize, log2 ratio, SES normalization but still get the same pattern as the plot below).

Any suggestion is welcome, thanks!!

[rls_08]

I have a similar issue with my data, but the baseline are remarkably different between my samples. what is the best approach for dealing with this if the goal is to compare occupancy levels across different conditions?

[ATpoint]

How did you normalize the data in the first place? This reads like your signal/noise levels are remarkably different.

Answer 1

[Devon Ryan]

There's no way to do this directly in deepTools, you'll need to save the data underlying the plot and fudge it in R (I have a feeling you're trying to over-interpret slight differences in ensemble enrichment over neighboring background between time points...be careful about doing that).

[chiefcat]

Thanks Devon. Yes, I also think it is dangourous to interpret the slight differences (I will try to convince my PI not doing that as well. I think it is much worth to spend time on searching real differences).

By the way, do you mean using --outFileNameData to save the data underpaying the plot? Which package can fudge the profiles? Thanks!

[Devon Ryan]

Yes, --outFileNameData, you can modify the data in base R (you're just adding/removing a small offset).

Answer 2

[Dave Gerrard]

I would suggest NOT creating lines on the same plot because for many of these kinds of differences it is hard to normalise out the library differences. In my opinion, and difference is better shown with two (or more) side by side heatplots running through the transcription start site. I've not used deepTools recently but know that ngsplot can produce side-side-side heatplots of two or more samples with the genes/transcripts in the same order on each (and optionally clustered and/or using ranks). These plots show better the co-variation across reference points and are relatively robust to differences in library size or distribution.

[Devon Ryan]

What you're describing is the default output of plotProfile and plotHeatmap from deepTools.