The ENCODE consortium published in 2011 a guide for RNASeq best practices (http://encodeproject.org/ENCODE/protocols/dataStandards/ENCODE_RNAseq_Standards_V1.0.pdf). They recommend "30M pair-end reads of length > 30NT, of which 20-25M are mappable to the genome or known transcriptome" for expression analysis and "100-200 M 2 x 76 bp or longer reads" for isoforms and novel alternative splicing discovery.

Additionally if you are thinking to use Tophat2/Cufflinks and use the mapped junctions, I DON'T RECOMMEND the 50 bp reads because by default Tophat2 splits the read into two segment to map them independently and analyze the intron-exon junctions (by default the mappers doesn't allow big gaps). The segment size by default is 25 bp. With a a read length of 50 bp probably you will have a read size ~40-45 bp after the processing (quality trimming), and in some cases will affect to the uniqueness of the read map (probably the segment size will be ~20 bp) (For an analysis of how the read size affect the uniqueness of the read mapping check the article from Storvall et al. 2013 (http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0053822).

Finally I don't know if it may help, but I have a presentation about RNASeq design that I use for my classes. You can take a look at: http://www.slideshare.net/aubombarely/rnaseq-analysis-19910448

Good Luck !!!

I am not completely sure for eukaryotes, as I do not know how repetitive transcripts in those are. But I would guess that this is more than enough. Or to put it the other way around: If you have a repetitive region of more than 50 bp, it is also likely longer than 75 bp.

In general, the total length of your sequenced cDNA pieces is more important in such cases than your actual read length. So the fragments should be as big as possible (600-1000 bp with linkers). Unless you are looking for small RNAs :-)

Usually 30-40 M paired-end reads (even 2x50 bp) are enough to identify differentially expressed transcripts, even in mammalian genomes. However consider that you may have variable rates of uniquely mapped reads. The longer is the read the better is the mapping procedure. You may have some difficulty to measure expression from lowly expressed genes and to detect differential abundance among transcripts from the same gene (originated by alternative splicing). However, if you aim to study only (or mainly) differential expression, your design should be ok, even if increasing the reads' number (50-60 M) would be desirable.

2x50 should be sufficient. For our RNAseq studies (prokaryotic and eukaryotic) we routinely ran SE 42bp runs and differential expression identification was not a particular issue. Note that the subsequent bioinformatic strategies used to analyse the sequencing data are as important as getting good sequence. Also decent sequencing depth, replicates and how the libraries were generated are also of importance.

Do you really need to sequence paired ends? With RNAseq you are really just counting transcripts, so there is generally no particular advantage in sequencing both ends of the same fragment (especially if you have a stranded library). With 100 cycles of sequencing to play with you might be better off doing single end 100bp reads than 2 x 50bp.

If you are using a MiSeq, the new v3 cartridges are very good. You could use the new v3 150 cycle kit and you will get a LOT more reads (25 million single end instead of 15 million single end with v2 chemistry). The cost is not much different and this will allow you to identify many more unique transcripts.

Good luck with your experiments :)

If your main goal is to determine differential expression (versus transcript structure), than your design should be sufficient. The paired-end reads alone will allow will increase your alignment fidelity tremendously.

The number of uniquely mapped reads will be crucial. Paradoxically, you can loose information when mapping on both sides. If your main interest is about differential expression, 2 x 50bases, or even 1 x 50 bases can be sufficient. The paired-end will become definitely important to assess splicing events.

I don't have any actual experience with eukaryote RNA-seq, but intuitively I'll go with that extra cost... If you look at Illumina page, they advice on using 1x50pb for gene profiling, 2x50-75 for discovery and 2x75-100 for complete transcriptome annotation. Also, you might seek for help at seqanswers.com. Those guys there know what they do.

2x50 should be enough. The most relevant parameter may be library preparation. For example, fragment size so that you can make the most out of your paired-end to better distinguish between different transcripts of the same gene.

Hi Brian,

We did single-end 101 BP reads for RNA-SEQ experiment.

The reads that were mapped to the reference were decent with very good coverage.. This inturn we used to look at differential exon usage.

Good luck

It really depends upon the goal and how exhaustively one want to investigate the deferentially expressed transcripts. For discovery phase, standards call for ~200 M tags (PE 2x76 or 2x100) and is considered sufficient. However, for routine expression analysis, 40-50 M tags should be sufficient, at 2X50 read depths. Project goal determines what depth and read lengths are used.

Agree with Daniel and the others. 2x50 should be enough.

As always, it depends on your coverage across your genes of interest. But 2x50 should theoretically be completely sufficient for alignment. I'd agree with the folks stating that longer single reads would also work perfectly fine if you're only keen to identify differential expression.

The long read paired-end business is really useful for characterizing alternative and unique splicing events. You mention this, but I don't know if it's of specific interest to your study or if you're just focusing on expression. In my work, whenever I found differential expression, I generally also found differential splicing behavior in the gene of interest and/or in a related gene.

Agree with Gavin Wilkie, single end 100bp would be my choice, allowing up to 2 mismatches when anchoring to the human genome.

Depends on whether or not you have an annotated reference genome. With a well annotated reference genome, you can get away with single end reads. Pair end reads assemble better, so if your beasty doesn't have a sequenced genome, it will make a very significant difference. It's really all about your read count, and the repeat nature of your genome/transcripts. ideally, you want to make sure that you don't double count the reads, or rather ascribe the reads to the wrong gene. I'd check your counts (predicted expression levels) with some qRT-PCR.

Hi!

I think there is no difference in 20/50/100/150 bp read length or singel/paired end, if you only want to look at diff. gene. exp., if you want to look at splicing events, thats another story.

I think the common thing to do is to use 50bp single-end. 50 because its the smallest of the defaults in i.e. HiSeq 2000 and single end because its cheeper than paired, meaning you can buy larger libraries at the same price.

The clue is to use as many biological replicates as possible and get as many reads as possible. Since everyone has limited with money, 50 bp single-end is the way to go.

I agree with all the people suggesting you 2x50 bp. in my experience, with longer reads your improvement in detection accuracy isn't "justified" by the increase of costs so...50 bp should be a good solution. good luck!!

Coverage.

The ENCODE consortium published in 2011 a guide for RNASeq best practices (http://encodeproject.org/ENCODE/protocols/dataStandards/ENCODE_RNAseq_Standards_V1.0.pdf). They recommend "30M pair-end reads of length > 30NT, of which 20-25M are mappable to the genome or known transcriptome" for expression analysis and "100-200 M 2 x 76 bp or longer reads" for isoforms and novel alternative splicing discovery.

Additionally if you are thinking to use Tophat2/Cufflinks and use the mapped junctions, I DON'T RECOMMEND the 50 bp reads because by default Tophat2 splits the read into two segment to map them independently and analyze the intron-exon junctions (by default the mappers doesn't allow big gaps). The segment size by default is 25 bp. With a a read length of 50 bp probably you will have a read size ~40-45 bp after the processing (quality trimming), and in some cases will affect to the uniqueness of the read map (probably the segment size will be ~20 bp) (For an analysis of how the read size affect the uniqueness of the read mapping check the article from Storvall et al. 2013 (http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0053822).

Finally I don't know if it may help, but I have a presentation about RNASeq design that I use for my classes. You can take a look at: http://www.slideshare.net/aubombarely/rnaseq-analysis-19910448

Good Luck !!!

It depend's. You want to identify differential expression between transcripts or will only use one transcript per locus?

Bigger reads are best to separate multiple alignments in cases of transcripts of the same locus (they share some exons) and with genes with repetitive regions in nucleotides.

In my opinion, if you have money, choose the bigger reads. But, if you chose 50 bp you can write a very good paper too.

Regards

It's true that the sequencing cost will significantly decrease but mappability will be affected. A member of our team is trying to answer the same question by assessing how much will mappability be affected. I'll get the info for you.

Though it's better if reads are longer but 2x50 is sufficient for your purpose. For differential expression identification, most important things are biological replicates. Without that no statistically significant diff. can be identified.

I agree with Tanmoy, replicates are definitely a must in any study on differential expression. However, what you should also ask yourself is: which genes are you interested in? In case you are going to investigate lncRNAs, higher coverage pays off as your dynamic range increases and you are able to check lowly expressed transcripts. If you want to identify novel transcripts, longer reads give you an extra advantage, as they are more likely to be intron-spanning.

Best,

René

If you can get longer reads (100/150bp), this would be better but 50bp should be sufficient. Note that your coverage is deep enough. It can be worth to read this review:

Sims, D., Sudbery, I., Ilott, N. E., Heger, A. & Ponting, C. P. Sequencing depth and coverage: keyconsiderations in genomic analyses. Nat Rev Genet 15, 121–132 (2014).

As long as measuring diff-exp is your goal, a choice between single-end or paired-end and read length wont be a big problem, specially in case of mouse where you have a well annotated set of coding and non-coding gene models. In fact the number of biological replicates is the most important factor (as suggested before by René, Tanmoy, Amanda and Sindre) followed by the mapping strategy. I would consider the following before proceeding

- At least three biological replicates for each sample.

- Accept reads with upto 10 multi-mapping on the genome (only in case of paired-end sequencing). Check this paper http://www.biomedcentral.com/1471-2164/12/516

However it is a completely different scenario if you want to look for

- Novel unannotated liver specific transcripts

- Alternative splice variants of annotated gene-models specifically expressed in liver

In this case 75 bp paired-end can provide a better coverage and add confidence to downstream assembly of novel intergenic non-coding trasncripts which may overlap repeat elements. My final recommendation would be to use Bioconductor packages like DEseq2 or EdgeR for diff-exp analysis rather than relying upon cuffdiff from the tuxedo pipeline.

As an FYI: The TCGA project uses 2x50 RNA sequencing for differential expression, fusion detection, and alternate splice form detection with good results.

https://tcga-data.nci.nih.gov/tcga/tcgaHome2.jsp

75bp paired-end is better ... leaves you with options in using the downstream analysis pipeline ... if its one time investment then I would recommend 75bp PE or 100 PE. If its just identification of transcripts 50PE will also be enough ... however, just like recommendations above ... it will restrict your choice of analysis tools ... and ofcourse the replicates ... it will definitely pay off to have replicates

For ordinary gene expression analysis RNA-seq is generally badly suited. I recommend Massive Analysis of cDNA Ends (MACE) (see www.genxpro.de). Either 3'- or 5'-MACE will give a much better coverage of all polyadenylated transcripts than RNA-seq at much lower costs. With MACE 10M 50bp single-pass reads for a known genome (as e.g. human) are sufficient to cover even lowly expressed transcripts. Of course this does not cover alternatively spliced transcripts. However, withRNA-seq, this requires approx. 200M reads. MACE nicely detects differential polyadenylation, on the other side, better than RNA-seq. I aggree that biological replicates are definitely important. With MACE they are affordable...

1x50 nt reads, i.e. single-sided sequences, are sufficient for a standard expression analysis in a model organism. Longer read length or paired reads I would recommend only if you aim at complicated genes, like paralogous gene families with high sequence similarity, or at detailed genes structures, like certain mRNA isoforms (alternative splicing). This recommendation is in accord to the ENCODE guidlines cited by Aureliano.

One thing that may be important for your considerations: What mouse strain do you want to analyze? If it is very different from C57Bl/6, then your read mapping may become affected. And this might be a reason to increase the read length (possibly 75 nt) in order to keep your mappings specific.