Optimal read lenght for RNASeq

[cjohnson]

Has anyone here done any research on optimal read length for doing RNASeq?

I've been able to find one paper so far that looked at read length (Li et al). Suggesting that shorted reads <35 are best.

I'm sure like everything it comes down to your experimental question, but I'm looking for some general advice.

Thanks,

[jdanderson]

Hello cjohnson,

I am far from an expert on this matter, but to get the ball rolling on your thread:

In general, the longer the better and paired ends (especially) are helpful. I think the added value is greater resolution (and statistical power) of smaller genes and as well as lowly expressed genes. I'm sure cost is a concern for you, as it is for everyone (if not, then i envy you), so you may find that adding a little read length is a bit cheaper than running PE's. If you are using a core facility, however, you have to consider how much of the slide you are using (# of lanes) versus how popular a particular read length is with your type of run, PE or SE. I've heard that 40bp SE reads are the minimum you should try to do, and of course the more biological replicates the more powerful your analysis will be.

Hope this was some modicum of help.

Cheers,
Johnathon

[Michael.James.Clark]

I'm interested in this question as well. Haven't done RNAseq myself for a long time, and back then it was 36bp PE. Now I've read 75bp SE is pretty standard and am considering 100bp PE. Anyone have a solid answer?

[mnkyboy]

If it is for mRNA-seq profiling then you will do well with just 54 bp. If it is for whole transcriptome, junctions, and structure longer is better. PE obviously helps there. But you may not get PE support for directional libraries.

We like 76 directional for whole transcriptome but that is because at 100 we start seeing that we will sequence the adapters sometimes.

[cndewey]

I'm an author on the Li et al. paper that was cited, so I thought I would add my thoughts.

If your only goal is to do quantitation and/or differential expression, we've found that the number of reads matters more than read length once you reach a minimum read length of about 25. So if you have a choice between one lane of 50bp SE vs. two lanes of 25bp SE, you should go with the two lane option. Similarly, two lanes of 25bp SE is better than one 25bp PE run. Of course, if you only have one lane to work with, having longer reads won't hurt, unless the number of valid reads goes down.

Another group has confirmed our results on this:
Nicolae et al.

If you want to do reference-based or de novo transcriptome assembly, then longer reads or PE reads will probably be more useful.

[malachig]

I agree with others here that the greater the length the better and paired is more desirable than single end. For paired-ends I wouldn't go shorter than 40 and for single end, no shorter than 50. I also agree that depending on application, depth may be more important than read length.

An obvious consideration related to read length is mapability of your reads. If you want to have some numbers to justify your choice you can download the 'mapability' tracks from UCSC and consider how many bases of the transcriptome (i.e. exonic bases) are uniquely mapable at various read lengths. For hg19, this has been pre-calculated for 24-, 36-, 40-, 50-, 75- and 100-mer sequences with up to two mismatches allowed.

Another suggestion. There are a number of caveats associated with comparing libraries of different length. So whatever you decide for length and paired vs. un-paired, stick with that for all libraries if you can...

[honey]

I was wondering what parameters should be optimal to get the out put in TopHAT.
Thanks