A really bad library?

[FrSr]

Hi guys,
I am a beginner in the data analysis of experiments RNASeq.
I am currently trying to align reads (36 bp) from an RNASeq experiment done on a machine Illumina GaIIx.
As a first analysis I am aligning the reads to the genome with Bowtie, in order to understand how many reads mapped outside coding regions of genes (theoretically, I do not expect to see anything).
For many samples I can map more than 70% of reads, but of these 60% -70% mapped out of the genes annotated regions.
Among the readings that map in the genes, many (60-70%) are potential "PCR duplicates".

Evidently there was some serious problem in the construction of these libraries and in fact I do not think they are usable for expression quantification.

So, these are my questions:
1) What is an acceptable percentage of reads (from a good library) that map on the genome outside regions annotated as genes?
2) Based on your experience, from SE library of 36-bp, what might be an acceptable percentage of "PCR duplicates" (I saw on the forum that this is a much debated topic...)?
3) In the case of a read that can be mapped in "n" places on the genome, above which the value of "n" is advisable to discard the reading (in practice I refer to the options -m/-k of Bowtie).

Thank you for your support!

Francesco.

[NextGenSeq]

It really depends how you make your library. If you use poly-T selection you will see more reads mapped to exons. If you use ribo depletion and random priming you will see a lot of transcribed non-coding reads not mapping to exons.

[steven]

I would say 60-80% of mapped reads in exons is pretty common (with polyA+ selection protocol). It also depends on the pre- or post-alignment filtering (are multi mappers kept, what is the sequence quality threshold, how many mismatches are allowed, etc). And of course on the reference annotation and species: for instance, RefSeq annotation describes reliable gene models but is not exhaustive, whereas Ensembl includes automatic gene predictions too.

Regarding PCR duplicates: how do you define them? I mean, how can you be sure that two identical reads come from the same mRNA and not from different molecules? The higher the expression level, the higher the probability to get identical reads from the same gene.. However, there is something you may want to look at, which is called "library complexity". Check this paper from Levin et al in Nat. Methods.