Why are Illumina paired-end SRA datasets made up of 3 FASTQ files?

[Bio.X2Y]

I'm looking at some NCBI SRA datasets for Paired-End Illumina Rna-seq.

In each case, the dataset is made up of 3 fastq files, even though I would only expect 2 (one for each end).

Example:

SRR018256.fastq (2,048,908 lines)
SRR018256_1.fastq (50,313,152 lines)
SRR018256_2.fastq (50,313,152 lines)

All files look OK, and the _1 and _2 files have the same number of lines, as I would expect.

Does anyone have any idea what the third file might be?

Thanks.

[Pepe]

What i do with my paired end reads is to filter out the ones that have adapters or bad quality. Then I take the pairs of the removed ones and I put them in a separate file, so the 2 paired end files have the same number of reads and in the same order but I can still use the 'pairless' reads in the analysis.
Maybe they did the same?

[Bio.X2Y]

Thanks Pepe, that makes sense.

Does anyone else have other possible explanations? Cheers

[Chipper]

Unpaired reads.

[Bio.X2Y]

Hi, is it possible to get unpaired reads from a paired-end experiment? I'm not very familiar with the procedure.

[GW_OK]

I should think so. it is possible that something went wrong with one read or the other, leaving a lonely, unpaired read.

[simonandrews]

I'm not sure the Illumina pipeline can create unpaired reads. The basis for the sequencing is an initial identification of regions followed by tracking those regions to determine sequence. When you do a paired end read there is no separate cluster detection in the second read, meaning that you use exactly the same regions as the first read.

For the output from the pipeline you get only two sequence files, one for each read, which always contain the same number of sequences and always come in the same order so you can match up pairs of sequences. If stuff goes wrong you'll just end up with a bunch of sequences full of poly-N.

If the file is for unpaired sequences then it must have been something which the researchers created from the original data, as the pipeline itself won't create this.

Could it be a trial run before the main sequencing run? We do this routinely with our libraries - doing 10% of a lane with them to see if they look OK before going on to do a full run.

[Bio.X2Y]

Hi Simon, thanks for this. Out of interest, when you say you do 10% of a lane, how is this done? I'm not very familiar with the sequencing procedure itself, but I imagined it was an all-or-nothing, and you couldn't back out after 10%. Do you mean you are watching some results in real time (i.e. nothing to do with the GAPipeline), and making a decision to abandon if necessary after 10%? If you do abandon, does this mean the flowcell is effectively wasted? If that's what happened here (in the decribed experiment), would the authors have needed to run image analysis, etc. on partially complete reads? Wouldn't that mean that (a) they wouldn't have full length reads, e.g. they would only have 5 bases per read out of 50 potential bases and (b) they would still be paired? Thanks for your help!

[simonandrews]

Hi Simon, thanks for this. Out of interest, when you say you do 10% of a lane, how is this done? I'm not very familiar with the sequencing procedure itself, but I imagined it was an all-or-nothing, and you couldn't back out after 10%.

We still run a control lane on each flowcell because of the nature of many of our libraries. What we can therefore do is to mix in 10% of another sample alongside the PhiX and then extract out everything which doesn't map to PhiX at the end of the run to get a small scale view of the other library.

[spadejac]

No other explanations. Here is NCBI documentation about it:

SRR000001.fastq – Fragment library data, or unpaired mates from a paired library.
SRR000001_1.fastq – First mate sequence.
SRR000001_2.fastq – Second mate sequence in the submitted orientation.