Several possibilities for reads aligned against intergenic regions: sequencing error, mapping error, unannotated genes. For human genome, it's very likely the last one is true. The annotation of a genome could never be finished, as it could never be sequenced completely.

Are you looking at human? Not sure about your data, but you could try to correlate with the ENCODE project results. From the Nature ENCODE site:

"On average, for each cell line, 39% of the genome is covered by primary transcripts and 22% by processed RNAs. ".

Also from http://bionumbers.hms.harvard.edu/bionumber.aspx?s=y&id=103746&ver=2

"Using both whole genome tiling arrays and tagging, the Encyclopedia of the DNA element (ENCODE) pilot project has demonstrated that up to 93% of the human genome may be transcribed in multiple RNAs ."

Several possibilities for reads aligned against intergenic regions: sequencing error, mapping error, unannotated genes. For human genome, it's very likely the last one is true. The annotation of a genome could never be finished, as it could never be sequenced completely.

Enhancer RNAs are another possibility to consider for explaining bona-fide intergenic transcripts. However, as several commentators already pointed out, some candidate transcribed intergenic regions may simply lack proper annotation.

Your question is also addressed but perhaps not fully answered in Fontdevilla (2011) The Dynamic Genome: A Darwinian Approach (books.google.cz/books?isbn=0191620793) in the chapter named "The RNA machine in the genome" on page 25.

The chapter mentions "pervasive transcription into many, often overlapping transcripts" and the possible regulatory roles of the RNA molecules, some of which are already known.

This really makes sense...if you count the number of repetitive elements that contain their own promoters, such as ERVs, LTR and LINE elements in the human genome, some of which exist only as fragments, usually containing the promoter part, the genome is literally full of promoters. Most of these are probably silenced, still they could be the main reason of the "pervasive transcription", in my opinion. The situation may be slqightly different in different genomes/species.

Sorry, can't help it .. still thinking about this problem and partly talking to myself:

What bothers me as a bioinformatician, is how much of this "pervasive transcription" could actually be "apparent pervasive transcription", present only because reads from one of many interspersed repeat elements would map to other locations in the genome and the mapping software would take its chances and place the reads randomly all over the genome. Does anybody know the statistics of read mapping and genome sequence composition intimately enough to rule out such possibility?

RNA reads which are showing alignment in intergenic regions are probably miRNA or siRNA because these mostly comes from intergenic regions or intronic of genome which regulates coding regions og genome.

Matej, you could look at the size of the repeats problem by playing with the alignment parameters, and comparing the results. One option is to set a parameter in the aligner for 'unique alignments only', so reads which do not align uniquely to a single position are not aligned at all. If you then take the unaligned fraction, and realign using a more loose approach, switching off the 'unique' constraint, you could see what effect it has? The size of the problem will really depend on several factors - the level of identity between the repeats, the length of the reads, whether you have paired-ends and the rate of sequencing errors.

Thanks for you guys. It's very helpful. Another related question is what about the reads mapped to the intronic region of the genome? Any difference between the intronic mapped reads and intergenic mapped ones?

BTW, I am dealing with human genome.

Intronic reads could be due to 1) immature transcripts with introns not spliced out yet 2) unannotated exons which might represent alternative splicing, if the sequence and mapping quality are reliable.

Intronic reads could be due to intron-inclusion splice-variants.

The quality of the cDNA library is also of concern. With your library is not polyA specific it may include genomic DNA fragments as contaminnats. These genomic fragments will map randomly over the genome, but they do not reflect real transcriptional activity.

Both cases may apply and NCBI refseq is far from perfection as a tool to clearly divide coding and noncoding sequences. You could have contamination of genomic DNA and/or gene variants with cryptic exons, In these cases it is advised to look at the flanking regions and possibly find junctional sequences. Alternatively, target resequencing may help in clarifying the nature of the gDNA sequences.

Those reads could be derived from non-coding RNAs that was not yet annotated in your genome.

Intronic and intergenic mapped reads could be long non coding RNAs, most of them are unannotated and they will appear as new transcript in tophat/cufflinkins

For you understand more about lncRNAs I copy 2 articles showing the pervasive transcription in human genome

http://genomebiology.com/content/8/3/R43

http://www.nature.com/nature/journal/v489/n7414/full/nature11233.html

However intronic reads are complicated to work because as said above by the other fellowns they could be pre-mRNA, gDNA and unannotated exons which might represent alternative splicing.

There are various types of different regulatory RNAs. If you get a little into the litterature on this subject, you will find, that some types of regulatory RNAs include poly-A tails, whereas others does not. So dependent on the protocol used for library-prep, you can expect to observe different RNAs which are not mapped to exonic regions of yourgenome.

Furthermore, none of the trimming and mapping algorithms used for RNAseq are perfect, meaning that the quality of the genome and reads respectively can pose major challenges and affect the mapping quality. In particular if you are working with repetitive or hypervariable regions.

Finding reads that map to intronic and intergenic regions is part of the fun and speaks to the benefit of doing RNA-Seq over arrays. We may not know how to deal with them right now, but it is highly possible that they represent something that we know very little about at this moment.

Three reasons for reads mapped to intragenic region. 1. one type of alternative spliced RNAs, intron retention. 2. premature RNA with introns yet to be spliced out. 2. DNA contamination. 3. DNA contamination.

Reads mapped to intergenic region could because of non-coding sequences from RNA-seq.

If we are doing Ribo depletion we are exposed to all the RNAs except rRNAs, which will include all the pre mRNAs and other non coding RNAs, because of that we will get lots of intronic reads.

Even with polyA selection we can capture immature transcripts where introns are not spliced out yet along with mature mRNA. We will get less intronic reads with polyA selection as compared to Ribo depletion.

Also As human genome is still incomplete there will be many unannotated genes.