chimeric RNAseq reads on Ion Proton??

[ElizabethRoss]

Hi All,
I am noticing evidence of chimeric reads coming from the ion proton, ie, the reads are half my organism of interest (lets say human) and half something else (Half ERCC control).

When BLAST'ed to nt, half the read hits human, and half it hits an ERCC control fragment. This is not a one off, I see it in data generated in-house and also some public human RNAseq data that I downloaded.

Has anyone else seen this? Are you aware of any documentation on these chimers? Do you agree with my interpretation of the result?

The example read is from SRA:
>SRR902918.363 AF94H:00015:07753 length=142
CGCATAGTCGCCATTCACAGATTTGCTCGGCAATCAGTACTGAAAGGCTAAGGCAGGTGAATCACCTGAGCTCAGGAGTTCAAGACCA

Thanks for your help.

[danwiththeplan]

That particular study from the SRA used five platforms (Illumina’s HiSeqs, Life Technologies’ Personal Genome Machine and Proton, Roche 454 GS FLX, and Pacific Biosciences RS) so I'd be curious as to whether the other platforms are showing chimeras as well.

[ElizabethRoss]

I had a look at one of the Illumina libraries from that study also; I could not detect chimeras in the Illumina data. I think the issue may be something to do with the proton library construction, most likely during a PCR step.
My main concern is the presence of chimera in the library that are two piece of independent human RNA that are joined, as they will be less trivial to remove/filter/trim.
I will have a look at the other platform data, but it may take me a few days to find the time.

[Brian Bushnell]

I've seen chimeric reads on various platforms, including Illumina. I think it is important to always design analyses to be robust against potential chimeras, as they could be anywhere. But hopefully, they will tend to be random joins, such that a given chimera only occurs once (aside from PCR duplicates).

When doing things like scaffolding, it is never safe to join two contigs on the bases of a single read pair (and I'm not implying 2 pairs make it safe, either), exactly because of problems like chimeric reads. And RNA-seq data seems to be even more prone than DNA to have strange reads that are possibly artifacts, such as chimeras.