How to get barcode fastq with CASAVA

[Malfet]

Hi, all,

With the previous version of CASAVA, one first had to do conversion of bcl to qseq files and then convert qseq to fastq getting separate fastq files: one with reads (two for paired end runs) and another one for barcodes. Next step was demultiplex your data using all these fastq-files.

New version of CASAVA directly converts bcl files into fastq and perform demultiplexing in the same time. Is it possible to extract barcode fastq files? I have not found how to do it in the manual provided for bcltofactqconversion script.

The thing is that, results of my last sequencing run looks quite weird and most reads can not be assign to a specific library, so I would like to have a look on the barcodes separately and check the sequencing quality.

Will be very grateful for any suggestions.

[GenoMax]

You can check on the barcodes (they would not have the Q-score) in the fastq headers of the "undetermined" pool files (I assume your sequences ended up mostly in there). You would want to look in the "Undetermined_indices/Sample_lane*" directories under the "Unaligned" directory. Depending on SE/PE run, there should be one or two files in there.

[Malfet]

You can check on the barcodes (they would not have the Q-score) in the fastq headers of the "undetermined" pool files (I assume your sequences ended up mostly in there)

I see! That's so simple... but could you suggest any way to analyze the sequencing quality?

I can see by eye multiple mismatches, so one way would be to repeat demultiplexing procedure allowing 1-3 mismatches, but I also would like to better understand what happened and whether it was the sequencing problem or libraries or anything else.

[GenoMax]

Number of mismatches probably would depend on your barcodes. You can experiment with "--mismatches n" (or "--mismatches n,n" for 2D barcodes) parameter for the configureBclToFastq.pl command. You can generally recover additional data by allowing for 1 mismatch.

If your sequence looks otherwise good then one surefire way of running into problems with tag reads is because of overloading of samples. Beyond what you can recover by the above method your only option would be to re-run with a lower loading conc (if you are seeing lot of N's in the tag reads).

[Malfet]

If your sequence looks otherwise good

Not really. This is paired-end run. Normally forward and reverse reads look quite similar in quality. However, this time quality per base for the forward read looks pretty different from the reverse and it makes me suspicious. I attach the graphs for both reads...
Within the barcode sequences I don't see N - I just see the wrong letter. G instead of C. C instead of A and so on. One or two mismatches within each barcode. Let's see how much I'll recovery with --mismatches 2 options on...

[GenoMax]

Looks like something was going on with this run/lane for multiple cycles. I am surprised this data was released (unless you had asked for it to be released). Have you inquired to see if this was a lane specific problem or a more general run wide one?

[Malfet]

Looks like something was going on with this run/lane for multiple cycles. I am surprised this data was released (unless you had asked for it to be released). Have you inquired to see if this was a lane specific problem or a more general run wide one?

Not yet, I'll tomorrow. Something was definitely going wrong there...