question about read depth for splice variant detection

[zhourui519]

Dear All,

I plan to perform RNA-seq recently, but I am a newer to RNA-seq and I found it was hard to determine the read length and sequencing depth(read numbers per sample).

I use human cell lines and I want to detect alternative splice variants as well as differential gene expression, I will perform RNA seq using illumina 2000/2500.
So my question is how many read length and read number should I choose?

It looks the read numbers is the more the better for AS, but due to our limit budget, I want to find a cost effective way but can still generate a good result.

The core in our university recommend me run 1 sample per lane, so the price will be too high to be acceptable. Do I really need that read depth?
I saw some people suggested >100M pair-end for splice variants detection, that means 100M reads or 100M paired-reads? I was confused.

Right now, I plan to use paired-end 2x69, run 3 samples per lane on illumina Highseq 2500 RAPID run(about 50M paired-reads per sample), is this reasonable? I am really not sure!

Please advice me if you could, thanks!
Rui,

[woodydon]

What cell line are you testing?

One sample per lane is correct in general. ~100M pair-end reads = 1 sample per lane. I would say the minimum is 2 sample per lane because HiSeq 2500 should produce about 150~160 pair-end reads.

A recent guideline for this: https://genohub.com/next-generation-sequencing-guide/

Woody

[Brian Bushnell]

If you want to detect alternative splicing, then longer reads are better. The total number of bases sequenced is obviously important, but you could sequence 100 billion 25bp reads without yielding useful results.

I suggest the longest read length available with near-maximal output. HiSeq 2000 can generate 150bp reads, which are good enough. HiSeq 2500 can generate longer reads at a drastically reduced throughput which I would not recommend for your purpose.

If you have any interest in alternative splicing, do NOT use 69bp reads - use the longest you can get without sacrificing throughput. Furthermore - if you map to a reference genome, and wish to discover new isoforms without bias, I suggest you use an aligner such as BBMap that is purely de-novo and thus untainted by existing gene annotation files or motifs. For vertebrate RNA-seq you should include the flag "maxindel=200000".