circRNA- seq normalization

Question

[IP]

Hi!

I am working with RNA-seq data and I have identified circRNA on them. Know I want to do the differential expression analysis. I am wondering which kind of normalization should I perform on the circRNA I have identified to perform the differential expression.

As circRNA identification tools try to identify a junction read of the circularization, my guess is that there is no need to take into account the gene length for the normalization. As it is only taken into account the fusion junction read. Would it be correct just to use tags per million?

Thank you very much

[vivekbhr]

If you are using edgeR/DEseq the library size normalization should be enough.

Answer 1

[Kanne]

If your data is from (rRNA-depleted) total RNA and your read lengths and library fragment lengths are the same between the samples you are comparing, then tags per million is an option but this is pretty crude; the suggestion by Vivek Bhardwaj to use the normalisation values from EdgeR/etc is also appropriate and should be better. I would clarify though, these should be the normalisation factors derived from quantification of linear RNAs, not circular RNAs. This is because if you use normalisation factors derived from quantification of circRNAs only then you are normalising away any difference in the total circRNA production between your samples. It is a more reasonable assumption that the total production of linear RNAs will be the same between samples than the total production of circRNAs.

Note: If you are using a different data type (e.g. RNaseR or polyA-depleted) or your read and fragment lengths aren't the same, neither of these methods is really appropriate though...