The fact that RNA-seq "can identify unknown transcripts" relates to transcripts that haven't been annotated yet. If you look at Ensembl's genome annotation (http://www.ensembl.org/Homo_sapiens/Info/Annotation), you'll see that the human genome has roughly 20,000 coding sequences, but also thousands of non-coding sequences (small and long non-coding RNAs, pseudogenes, etc).

However, due to transcriptional and post-transcriptional (such as splicing) alterations, these genes may code to up to 200,000 transcripts. This number is likely to increase because there are many transcripts which have not been annotated yet ("unknown"). With RNA-seq you have the option to assemble your transcriptome from scratch ('de novo transcriptome assembly') and discover these potentially novel transcripts. This is not possible with microarrays, which contain a pre-defined number of probes (usually representing 20,000 coding sequences). Of course, this is a rough representation of the human genome, and it does not take into account different transcripts of a single gene. In this sense, RNA-seq provides better resolution than microarrays.

Serial analysis of gene expression, developed by Velculescu VE; Zhang L; Vogelstein B; Kinzler KW. (1995). Science. 270 (5235): 484–487, is an open method in the sense that it can analyse both known and unknown transcripts may prove to be useful in your gene expression analysis.

There is also the additional complication that there is a lot of features you will not capture with microarrays but you will with RNA-seq - the most prominent is you can find isoform switches with RNA-seq data.

In addition to differentiate between isoforms or yet unknown transcript as bonus for RNA-seq instead of microarray gene expression profiling are:

• identification of fusion genes
• calling for variants (mutations)

I agree Terezinha Souza's response. RNA-Seq will be the gold standard for identifying/quantifying novel/non-annotated transcripts. However, Microarray can not be excluded in some instances:

If you need further help you can contact me @ [email protected]

MedGenome offers RNA-Seq as well as array based transcriptome profiling.

In Response to Ravi's comment.

While I agree that for low-input studies RNA-seq are still challenging (it can be done and there are kits specificly designed for that) I strongly have to disagree with the number of reads Ravi suggest is needed for RNA-seq!

If you are just interested in differential expression on gene level you will get by just fine with using 10-20 million (single end) reads - meaning the cost is very low. Most studies I know of shows that this is the region where gene-level DE analysis starts to saturate.

If you want the transcript level resolution it is clear that you need to sequence deeper than for gene level - but currently most people are recommending 40-80 million reads not hundreds of millions!

That said because of the unbiased nature of RNA-seq you can naturally keep sequencing and you fill find new rare and/or unstable RNA molecules.

If your goal is to do differential expression analysis as soon as you get above the numbers mentioned above the limiting feature is the number of biological replicates you have. Therefore I would encourage you to use the money on replicates rather than extreme sequencing depth.

/Kristoffer

Kristoffer Vitting-Seerup: Please check the recommendations for RNA-Seq from ENCODE Consortium

Dear Ravi.

While I agree that more sequencing is better I'm speaking from a practical point of view - as I said you can keep sequencing and not reach saturation (no-one have to my knowledge...?).

For a gene level analysis it is clear that even ENCODE data can be subsampled quite a bit without sacrificing to much accuracy in the gene differential expression analysis (as long as you have 2+ replicates). See fx figure 5 in https://www.ncbi.nlm.nih.gov/pubmed/24020486 

For transcript level analysis the key in ENCODE's recommendation is the "which sensitivity of detection is important" meaning for instances where you want to identify the majority of transcripts present in a cell. In most practical settings we are not really interested in rare lowly expressed transcripts - only the major transcripts within genes - whereby we can do with less sequencing.

The most important consideration is however that if the goal is differential expression analysis the number of replicates you have very quickly becomes the limiting factor so money are better spend on biological replicates than sequencing depth. One of the better analysis of this problems can be found here: https://www.ncbi.nlm.nih.gov/pubmed/23497356 (incl supplementary figures).

/Kristoffer

That's what I wrote in regard to ENCODE's recommendation: "sensitivity of detection is important"  to identify the majority of transcripts present in a cell, which requires very high depth of sequencing  (>100M and in some instances >300M). It seems you agree to this point.

I suggested to decide based on the objectives of the study (whether to quantify expression changes or to detect novel transcripts and following functional studies).

Please go though my final suggestion in my previous reply.