PE or SE reads for alternative splicing / isoform quantification?

Question

[nevev]

Hi All,

I've had a long break, and last time I learnt about it, I thought PE reads are required for detection and quantification of differential isoform expression. A few days ago someone was trying to convince me that long SE reads (100-150bp) can do as well. Any comments/ideas/opinions?

I am planning an RNA-seq experiment now, and need to decide if I want to pay 2x for PE, or can settle for SE. My main interest will be in DE of genes, but if I already do it, I'd rather have a dataset that'll allow me to look at isoform expression, TSS usage etc. too later on (if I recall correctly, CuffDiff could do all that).

Thanks in advance for good advice,

n

Answer 1

[Brian Bushnell]

SE reads are absolutely capable of differential isoform quantification. The best example, of course, would be PacBio reads-of-insert, which give you whole-isoform reads with fairly low error rates (the error rate depends on the isoform length). For Illumina reads, you have two choices:

1. Merge overlapping paired reads, then map them as single-ended, using a splice-aware tool such as BBMap; then, use any tool to identify isoforms.

   or

2. Map the paired reads (again using a splice-aware aligner) and rely on post-processing to identify isoforms via a pair-aware protocol.

PE is strictly better than SE for isoform identification, whether or not you decide to aim for overlapping reads that you merge prior to quantification. But it PE is not required, particularly with longer reads.

It's important to decide whether you care more about isoforms or genes. For gene expression quantification, more fragments are better; for isoform expression quantification, longer and paired reads are better.

[nevev]

Thank you for the answer!