Makes sense to conduct SALMON mapping followed by NOIseq for differential expression analysis?

Question

[YEODYNASTY]

Dear all,

For my RNA-seq project, I'll like to use SALMON to obtain GC bias-corrected count table from my fastq files.

Since SALMON does not conduct differential expression analysis, I'll like to confirm that it is acceptable to use any approach for the analysis, such as NOIseq.

Is it also OK to use any normalization method, such as FPKM and TMM, subsequent to SALMON?

I know the questions are very basic, but any comments will be appreciated!

[YEODYNASTY]

Hi, I just want to clarify with SALMON 1.3, we can now use the --geneMap flag to quantity counts at the gene level by providing the gtf annotation file.

Thus, there is now no need to sum up over the transcripts?

[ATpoint]

This option is quite old actually, I never used it, nor ever will since tximport is my standard pipeline. Cannot comment on that option.

Answer 1

[ATpoint]

Since salmon produces transcript level counts you should first sum these to the gene level, e.g. tximport from Bioconductor. From there on you can use any gene level DE framework you like. I personally use edgeR, but DESeq2 and limma are fine as well. Never used NOIseq. Please don't use FPKM, the typical DE tools expect raw counts anyway, please read the manual of e.g. edgeR and DESeq2.

[YEODYNASTY]

Thank you for your prompt reply.

You made an interesting point to sum transcript expressions to obtain gene level expression.

Can we provide reference gene sequences to SALMON, instead of the reference transcriptome? In this way, we can skip the summing-up step?

[ATpoint]

No, salmon quantifiers against a transcriptome. tximport is not computationally-expensive and it is a one-liner basically, takes like seconds to complete the process towards computation time, give it a try :)

[YEODYNASTY]

Noted, with thanks. :)