Quantifying contamination in RNA-Seq

[ruggedtextile]

I have some RNA-Seq data with a low proportion of reads aligning to the human genome (<50% in most samples, <20% in one or two). The sample prep involved very low numbers of human cells and so we are wondering if bacterial or fungal contamination could have occurred which may explain this.

Does anyone know of a resource/technique whereby I could test for this contamination? At the moment I am downloading bacterial 16S ribosomal sequences from RDP to see if I can align reads to those, but I'm wondering if there isn't a better way.

[bruce01]

Have a look at FastQ Screen from Simon Andrews: http://seqanswers.com/forums/showthread.php?t=10288

[ruggedtextile]

Thanks! Looks good.

[bruce01]

Yes it is! Just make sure you have all potential contaminants bowtie indexed.

[TonyBrooks]

How did you prep your samples? If you only had a few cells to begin with, I'm assuming there's some kind of amplification process going on during library prep.
Run your data through FASTQC. It should pick out any over-represented sequences (i.e. amplification primers).
We've done low cell number RNA-Seq before with the SMARTer kit from Clontech. It's good but there's a large amount of primer in the sequencing. Getting rid of these sequences improves our alignment %'s massively.

[ruggedtextile]

We used the NuGen Ovation kit. I did see quite a bit of primer contamination (detected with FASTQC), but even after trimming that out with reaper I still see a very low alignment %. Still trying to work out where the rest of the reads are coming from...

[TonyBrooks]

I take it you are aligning to the genome as you said and not transcriptome.
We've had problems with DNA contamination when we tried NuGen preps. We had a lot of reads mapping to non-exonic sequences (due to the random priming).

[ruggedtextile]

Yes, aligning to the genome.

I get poor levels of aligning (<50%) and then within the reads that do align very high levels of ribosomal, intergenic and intonic sequences, so in the end the fraction of usable reads for transcript estimation is too small.