Regarding TMM normalization before DEG analysis by DEGseq

Question

[seta]

Hi all,

Please kindly advise me if using TMM method for read count normalization and then DEGseq for identification of differentially expressed genes is the right approach? I never this tool, but as I searched this package shouldn't be used for the analysis because its statistical assumptions are wrong. Could you please provide me some references for it?

Thank you

[andrew.j.skelton73]

Check out DESeq2

[seta]

I asked about DEGseq, it differs from DEseq.

[andrew.j.skelton73]

Ok. According to the vignette the DEGexp function expects raw counts, as such TMM normalised counts breaks that assumption. The reason I suggest DESeq2 is that it's a very well established package for analysing RNA Seq counts.

I never this tool, but as I searched this package shouldn't be used for the analysis because its statistical assumptions are wrong. Could you please provide me some references for it?

Where did you find this information?

[seta]

Thank you for your response. As far as I know, TMM is between-sample normalized count (not break the assumption that you mentioned, please correct me if I'm wrong), and, therefore, more suitable for DE analysis, do you agree with me? However, DEseq or edgeR has internally calculated TMM, but unfortunately, it is not clearly specified about DEGseq, have you any opinion about normalization way in DEGseq?

As I search, DEGseq assumes a Poisson distribution, whereas edgeR and DEseq2 assume a negative binomial (NB) distribution that is more appropriate than the Poisson distribution for an RNA seq analysis. Do you agree with that the Poisson distribution assumption is not suitable for RNA-seq analysis?

Answer 1

[theobroma22]

You can use Poisson for RNA-seq. The NB distribution is more widely accepted cause it performs quite well when controlling for the low level transcripts. Either one could be used and are accepted for DE analysis.