Hi Everyone,

I have mouse Short-read RNAseq data set for treated and normal tissue samples with five replicates in each group. My goals are to identify Differentially Expressed Genes and Isoforms. I have performed read mapping with STAR and performed RESEM transcripts quantification. I have read about widely used tools for the statistical modeling of differential expression analysis (gene/isoform levels), Bioconductor R packages: DESeq2, fishpond; And workflows: rnaseqDTU, rnaseqGene for differential expression analysis.

DESeq2 is widely used differential expression software. and i guess it can be easily adapted in my workflow pipeline after converting my counts from RSEM by "tximport". However, i don't know it's capabilities for differential isoform analysis. can you please guide which software combination i should use for differential gene/isoform analysis with which normalization strategy?

I also read rnaseqDTU and fishpond both can perform differential isoform analysis. For rnaseqDTU, following Salmon quantification this workflow uses Bioconductor packages tximport, DRIMSeq, and DEXSeq to perform a DTU analysis. It also shows how to use stageR to perform two-stage testing of DTU, a statistical framework to screen at the gene level and then confirm which transcripts within the significant genes show evidence of DTU.

Which is the preferred software the fishpond package or rnaseqDTU after salmon quantification? can fishpond screen at the gene level and then confirm which transcripts within the significant genes show evidence of differential isoforms? can i proceed with RESEM and DESeq2/fishpond both? In differential transcriptomic analysis with the above mentioned tools which normalization strategy is suitable (TMM/size factor analysis)?

I saw at some places others used DESeq2, and fishpond both and cited in the manuscript (https://www.nature.com/articles/s41467-023-39945-w.pdf) , are the two softwares have their own pros and cons? Is it recommended to try different statistical modeling approaches for comparable differential gene expression results?

Thank you for taking sometime to read this.

Regards

We recommend fishpond's swish method for isoform-level DE analysis as it takes into account inferential uncertainty that is present when you use short reads. It has a vignette showing how to use it. We recommend using a method like Salmon for quantification, and then you would specify a number of bootstrap replicates (--numBootstraps 30). I will note that edgeR also has recently published a method (2024) which uses inferential uncertainty from upstream quantification tools.

Which is the preferred software the fishpond package or rnaseqDTU after salmon quantification? can fishpond screen at the gene level and then confirm which transcripts within the significant genes show evidence of differential isoforms?

rnaseqDTU is not a method but a workflow showing different methods. It is also not for differential expression per se. A gene which changes it's expression level will not necessarily be detected by the methods in rnaseqDTU, because those methods are designed for detecting switching of isoform usage within gene.

You could use fishpond::swish() for both gene and isoform level (see vignette). You can screen at gene level and then confirm which transcripts are DE, using the stageR method on fishpond's p-values.

But first I think you need to decide if you are interested in gene/isoform DE, or in DTU.

which normalization strategy is suitable (TMM/size factor analysis)?

The methods all incorporate their own normalization, and the methods are actually quite similar in practice (TMM/median ratio).