Tn5 bias in atac-seq

Question

[srhic]

Hello,

I recently got back reviews for a paper that contained some atac-seq analysis. The paper followed a standard analysis pipeline where peak calling was done using macs followed by differential analysis between two samples using deseq2.

One of the reviewers mentioned that we should include 'an analysis accounting for Tn5 cleavage bias' and that this can be done 'by analyzing nucelosome positioning at different locations in the genome'. Now I am a little confused by what this actually means.

I have found nucleosome calling pipelines like nucleoatac and hmmratac and can make nucleosome positioning plots but I am not sure how that relates to Tn5 bias. I was wondering if anyone here has any ideas on what kind of analysis can be done illustrate the absence of tn5 bias?

Thanks

[ATpoint]

The chromVAR paper has addressed the Tn bias in ATAC-seq but in a motif- and singlecell-related context (the latter towards how the bias correction improves the separation of cell types in UMAP/clustering if memory serves). It is (at least to me) questionable whether the Tn bias indeed makes a notable impact on DE by increasing power for regions that are preferred by the Tn and vice versa. I have not seen a paper addressing this in a DE context yet. ChromVAR implements a bias correction strategy but I am not sure whether the corrected counts can be used directly with GLM-based DE frameworks. Realistically I do not think that the overall biological message will change. You will probably gain a few differential regions which are close below the significance threshold and lose some that are slightly above but this (probably) will only be true for regions with low counts so on the very left of the MA-plot. Again, questionable whether this really affects whatever your paper has as main message.

I would do a simple check and see whether the bias correction indeed has a notable effect: Maybe use limma-trend on the log-counts and bias-corrected log-counts (from chromVAR) and hope that the results are almost the same (or not) I don’t know in fact.

[i.sudbery]

An alternative might be to do a motif enrichment analysis on your DE peaks, and see if you get the Tn5 preffered sequence coming out as a motif.

[srhic]

Thats a good idea. Thanks

[srhic]

Thanks, I will try the chromVAR method. I also expect that accounting for Tn5 bias will not significantly change my results. Based on some quick reading it seems to me this is more of an issue in case of transcription factor footprinting or motif enrichment etc. Since all I have done is a simple differential analysis of all atac fragments (both nfrs and nucleosomal), I dont think tn5 bias would affect my results.

I am still confused by how nucleosome positioning plots could address this question though.

[ATpoint]

The question is whether the reviewer really has detailed insight into this or whether he/she simply picked up some buzzwords in a paper. Do you have evidence based on the other comments that the reviewer is a computational person who is really an expert in this? If not I think you could satisfy them by showing confincing arguments against global effects of Tn bias.

[i.sudbery]

I guess all ATAC experiments will have some Tn5 bias in them. If you are looking for differential peaks, I guess this would show in having a higher power to detect difference where there are more, closer, matches to the Tn5 consensus. What this has to do with nucleosome positioning plots I'm not quite sure.

[Hughie]

Hi srhic,
I'm interested in this question about Tn5 bias.
The Tn5 indeed has some cut bias as reported, however, it seems these analyses mainly focus on footprinting. How the bias correction affects the peakcalling or differential peak is really interesting. I'm wondering if you can share your results about this part:

"I will try the chromVAR method. I also expect that accounting for Tn5 bias will not significantly change my results."

Besides, the question about nucleosome positioning and Tn5 bias is also interesting. I wonder whether this reviewer means correcting Tn5 bias can affect the nucleosome positioning prediction?
As I know currently, only nucleoatac and hmmratac (as you mentioned) can do the nucleosome calling for ATAC-seq data, however, they did not consider the bias correction procedure. So, I wondering how you replied to this issue.

Many thanks in advance.