I'm looking at ways to detect differential expression of antisense transcripts from stranded RNASeq data. So my first question is: Are there any Bioc packages designed for this type of analysis? I couldn't find any obvious...

Obviously one can just count up antisense reads over gene bodies and perform the usual DE analysis using DESeq2, edgeR, limma/voom, etc... Doing this however has a tendency to simply find genes that were proportionally DE on the sense strand as well. I.e. antisense transcription simply goes up linearly as a function of sense transcription. I am more interested in the cases where there is DE of antisense without (or even in opposition to) DE of sense.

One idea I had was to hack DEXSeq into treating the antisense strand of each gene as a pseudo-exon and then use that workflow to detect 'differential antisense use' - as opposed to 'differential exon use' as DEXSeq normally defines. Does anyone have any comment on whether that is a reasonable approach?

Dear Alex

sounds very reasonable. Simpler even may be to follow this approach: DESeq2 testing ratio of ratios (RIP-Seq, CLIP-Seq) where you basically think of the 'sense' and 'antisense' count for each gene as two different 'assays' on that gene.

As Steve says, it will be good to pair this with extensive visualisation to distinguish e.g. changes in transcript length from ones in expression strength

Wolfgang

Sounds reasonable to me ... as long as you have stranded RNAseq data, of course.

I'd guess you might have to refine your definition of what the "unit" of the antisense transcript is (ie. do you really want to treat the whole antisense region of the entire gene as one unit(?)), but aside from refining some of these things, I think using DEXSeq for this would be a good start!