Inverted TSS Enrichment Plot

Question

[nmalexan]

Hello,

The following is code I used to produce a heat map of my ATACSeq data. The strange thing is that I have one replicate that looks normal and another that looks like this (image attached). The code is run the exact same way for both samples.

bedGraphToBigWig Sample_treat_pileup.bdg genomic.fna.chrom.sizes Sample.bw

computeMatrix reference-point --referencePoint TSS -b 3000 -a 3000 -R genomic.gtf -S Sample.bw --skipZeros -o Sample.computeMatrix.gz -p max/2

plotHeatmap -m Sample.computeMatrix.gz -o Sample.plotHeatmap.pdf

I am getting a TSS enrichment that look inverted

Has anyone run into this pattern before? It doesn't seem biologically plausible.

[nmalexan]

For comparison, here is my technical replicate that looks normal. Keep in mind the first sample (with the inverted heatmap) had a tapestation without periodicity, whereas this second sample had periodicity.

[GenoMax]

here is my technical replicate that looks normal

Technical replicate or did you mean to say experimental replicate? A technical replicate (e.g. same library sequenced twice) should (will) never look different.

[nmalexan]

Yes sorry I meant experimental replicate

[nmalexan]

I'm getting a distribution that looks a bit like a camel hump when I zoom out

[rfran010]

Yes, this is strange. Even with a poor library, I don't think I'd expect a TSS depletion. It seems you are using macs2/3 generated signal files? Is it possible you are using some sort of normalized file (e.g. signal over lambda)?

[nmalexan]

Macs3! And I made sure the treatment bed graph is being used. Doesn't seem like theres any normalization happening but according to the paper I post below maybe it can happen??

[benformatics]

I haven’t looked at ATAC-seq data for almost 10 years but is it possible that you overdigested (in theory the TSS is open) and then during size-selection the really small fragments were excluded?

Other than that maybe look at your peaks from known good genes in IGV.

[nmalexan]

This is what im thinking too

Answer 1

[ATpoint]

My take on this, doing this assay for almost 10 years is to simply look at the IGV tracks. These diagnostic plots to me have very limited value. Use either bedtools genomecov or deeptools bamCoverage (with a bin size of 1) to make a bedGraph/bigwig, and then load into the IGV. Check whether there is good separation between peaks and noise. It is relatively easy to judge imo since separation should be obvious. Also check FRiPs, so fraction of reads overlapping peaks versus non-peaks. That will give in my hands enough indication whether data is worth following up on.

[nmalexan]

From this paper, im seeing the red pattern in figure 1C:

Yan, F., Powell, D.R., Curtis, D.J. et al. From reads to insight: a hitchhiker’s guide to ATAC-seq data analysis. Genome Biol 21, 22 (2020). https://doi.org/10.1186/s13059-020-1929-3

It seems that this can happen when there is overtagmentation and mononucleosomes are enriched relative to NFRs.

I agree, ill do what you suggest and take a look!

[ATpoint]

Check the bioanalyzer of the library. Overtagmentation is a thing to check before going to the sequencer.

[nmalexan]

So here's the tape station, I only have the one peak with no periodicity suggesting overtagmentation. The ActiveMotif website describes peaks like this as acceptable, but in my experience I rarely get good data unless there's at least a secondary peak.

[ATpoint]

This is clearly a failed experiment. I would not even have sequenced this. The banding pattern is THE quality control that must be fulfilled as it is the key readout towards the principle behind the (successful) assay. Sorry, not what I you wanted to hear probably...

[nmalexan]

That's actually exactly what I want. Can you tell me what you are referring to specifically in the tape station.

I want to know what to look for.

[ATpoint]

There should be a notable banding pattern with the individual bands representing nucleosome-free, mono-, di-, tri and so on nucleosomes, see original ATAC-seq paper from 2013. A single peak indicates failure. Something like this should be:

[nmalexan]

Excellent. Just to confirm sometimes I will have a peak like I had before but with some subtle secondary and tertiary peaks. These seem to yield good data. I assume these will suffice too, although the situation you post is optimal.

[nmalexan]

Another example