RNA-Seq quantitation for human, paired end vs single end, strand specific or not?

[Justin AC Powell]

Is there any consensus of opinion on the best RNA-Seq approach for quantitation of alternative splicing on well annotated genomes (e.g. human)? We don’t want to look for novel splice isoforms, we want to detect regulation of alternative splicing by our treatment among the annotated isoforms of a gene. For roughly the same price I can get:

60M x 50 bp single end reads

30M x 50bp paired end reads random primed cDNA

30M x 125 bp paired end reads using strand specific cDNA library

I guess in part this depends on the differential splicing analysis method chosen – something that counts only exon inclusion should work just as well with single ends. If you look for exon exclusion then you need reads that map across splice junctions, I don’t know if the chances of that are improved more by increasing the read length, or increasing the read depth (naively it would seem doubling either would have the same effect). Finally I don’t know what the benefit of strand specific libraries are for quantitation, one concern is that if they are more involved to produce then they might introduce variability - obviously in some cases they will help discriminate between sense and anti-sense transcripts from the same region.

[Bukowski]

'Of a gene'? A specific gene? How highly expressed?

Ever considered the Iso-Seq PacBio protocol?

[GenoMax]

For reference (cross-posted): https://www.biostars.org/p/173399/

[Justin AC Powell]

It's exploratory - so we're looking genome wide. Our vendor doesn't seem to recommend Iso-Seq for differential RNA expression but I haven't discussed why.

[danwiththeplan]

While trinity is obviously not relevant for humans, this vid explains strand specficity quite well:

https://www.youtube.com/watch?v=n2XEsw7EJLw

You will likely miss antisense transcription without it, or worse, mistake it for sense transcription, which could cause issues resolving isoforms?

[gringer]

No eukaryote genome is sufficiently annotated that novel isoforms won't be found in RNASeq data. Even if that were the case, some genes have fairly complex splice variants that benefit a lot from strand-specific sequencing. As an example, the transcriptome annotation shows genes that overlap with another gene on the opposite strand, and resolution of these genes will be impossible without strand-specific sequencing.

Regardless, given the choice between "60M x 50 bp single end" and "30M x 125 bp anything", I would choose the longer reads. Even if it were single end, you're getting a higher coverage with the longer reads (60 * 50 vs 30 * 125).