RNA-seq in the NovaSeq

[scrosby]

The smallest S1 kit is 100-cycles. With RNA-seq this results in a choice of single-end (100bp) or paired-end (2x50).

For the detection of splice junctions in, say, 300bp fragments...

• Perhaps there is a lower false positive rate when you have actual evidence of the junction in a read, which would argue for SR.

• Perhaps more (~3x as many?) junctions can be detected through a combination of direct detection and imputation by doing PE reads.

Has any definitive work on SR vs PE RNA-seq been published? I looked around a bit and found some chatter, but nothing peer-reviewed that directly addresses this question.

[scrosby]

260 views...NO replies!!
Maybe a more extensive search would have revealed nothing!

[Markiyan]

I would go for PE reads in this case.

Assuming yours RNAseq libraries are compatible with patterned flowcells:

https://sequencing.qcfail.com/articles/?report=reader

https://sequencing.qcfail.com/articl...ted-sequences/

and you use dual unique barcodes for demultiplexing to reduce index crosstalk, I would go for 2x50bp reads.

But given you probably would like to also see novel splice junctions in addition to annotated ones one may go with longer reads - use a 200cycle or 300cycle kit in PE mode, than use FLASH or similar tool to preassemble paired reads, and this would give you both "SR"(combined) and PE(non-combined) reads from yours 300bp library.
The longer the reads, the higher the "SR"(combined) proportion would be. and the more resilient it should be to the repeats.

PS: if you can find a nextseq 2x150 bp RNASEQ experiment in the SRA, than you can generate both 1x100 and 2x50bp reads from it in silico by trimming, and than compare the results after analysis by your pipeline.

PPS: If you have 200 or 150 cycle S1 kit - run it as 2x100 or R1:100+R2:50 and generate both 100 SE and 2x50PE datasets from the same run by doing custom trimming. This should give you a very good idea which method would be the best in your case for future experiments.