PE vs SE for RNA-Seq

[mike.t]

I have used paired end sequences for RNA-Seq experiments in the past with the understanding that PE reads are 'more accurate' somehow. One paper that I found says that the false positive rate is lower with PE. Single reads are a lot less expensive and I may be able to afford to sequence the libraries deeper. What do others on this forum use? PE or SE for rna-seq?

[dpryan]

For normal differential expression single-end 50 base reads are sufficient. We tend to use paired-end reads whenever we need to look at splicing, though, since inevitably that works better.

[SylvainL]

Actually, if the organism you work with is danio rerio (zebrafish), it may help to use paired-end even for single DE (a lot of duplications in this genome). For mouse or Humans, single-end are fine...

[prathima]

Paired-end sequencing facilitates detection of genomic rearrangements and repetitive sequence elements, as well as gene fusions and novel transcripts.

[Brian Bushnell]

I would always recommend PE for anything other than quantification studies. For quantification, it depends. But otherwise - per base, PE should always be cheaper than SE, and give massive advantages for accuracy.

This:

Single reads are a lot less expensive and I may be able to afford to sequence the libraries deeper.

Should, in the general case, be false. I guess it depends on how you define "deeper", though. Normally I think of it in terms of fold coverage rather than the total quantity of unique fragments.

So, it depends on the goal of your RNA-seq experiment! If you want to identify isoforms, or want maximal specificity, absolutely you should do PE. If you are just interested in differential expression of genes in general and don't care which isoforms are expressed, then SE will be more cost-effective.

[MikhailFokin]

Sorry, may be a bit off-topic. I personally use PE libraries, but just recently found the issue with read-through (when forward and reverse reads overlap completely). The question is - to drop reverse ones or leave them? Could it bias coverage and downstream DE analysis? If drop, then how to deal with less overlapped pairs? Ideally I would like to preserve non-overlapped fragments only, but not sure if there is a software capable for...

[Brian Bushnell]

If you preserve non-overlapping fragments only, you will get a bias against shorter transcripts. So I don't suggest it. Though if you really want to, you can do so by adapter-trimming and setting the minimum length to the initial read length (filtering out anything with adapter sequence).