Per tile sequence quality - when to go back to the sequencers?

[mlewis]

Hello,

This is my first time posting, but I have had a lot of help from the SEQanswers forums while browsing anonymously in the past, so thanks to everyone who has asked and answered questions on here.

We recently got some RNA-seq done by a commercial company, and I ran FastQC to check the quality before adapter removal and trimming. We submitted 36 samples for paired-end sequencing, 150bp, on an Illumina NextSeq 500. About 11 of our 36 samples had per-tile sequence qualities like the attached, with smears of poor quality in four places. All of them have the poor quality line around tile 21409. This is only the second time I have handled RNA-seq data, and the first batch was from a different centre and was much better quality (at least with respect to the per-tile sequence quality). My question is, is this something I should raise with the company who did the sequencing, or is it typical variation that won't cause any trouble further down the line? I am planning to tidy the data with Trimmomatic before aligning.

Many thanks in advance.

[Brian Bushnell]

You should go back to the company in one of these cases:

1) The data is so bad you can't use it without compromising your experiment.
2) The data is usable, but bad enough that it might affect the quality of your experiment, and it violates the minimum standards specified in the contract, and the time/cost on your end of getting new data is less significant than the what you expect to be the impact of the reduced quality on your results.

I don't think that data looks particularly bad, but it depends on what you're doing. I encourage you to try out FilterByTile, though. NextSeq is not a super-high-quality platform and if you get sequence data from a NextSeq you can't hold it to HiSeq 2500 standards; no matter how many times you re-run, it won't be perfect.

[GenoMax]

I concur with @Brian. You should analyze the data and then see if there are any particular issues noted with the results.

[mlewis]

NextSeq is not a super-high-quality platform and if you get sequence data from a NextSeq you can't hold it to HiSeq 2500 standards; no matter how many times you re-run, it won't be perfect.

Ah, that's interesting - I didn't know that. Thanks very much for the advice.

[Brian Bushnell]

Ah, that's interesting - I didn't know that. Thanks very much for the advice.

Illumina knows this, but they do not like to broadcast it, since they are trying to kill off their older platforms and sell newer platforms. So you will not get this kind of information from their official literature

[mlewis]

Illumina knows this, but they do not like to broadcast it, since they are trying to kill off their older platforms and sell newer platforms. So you will not get this kind of information from their official literature

Well naturally! Something to keep an eye on for future projects.

Do you know why there's been a reduction in quality, given that it's a newer platform? Is there a new high-quality one available as well as the NextSeq?

[Brian Bushnell]

NovaSeq looks promising. I'm not done analyzing the data yet; for our first run read 2 had lighting failures, but read 1 was very accurate. It has other issues, though, like barcode crosstalk.

In my testing, Illumina's two highest-quality platforms are still HiSeq 2500 and MiSeq.

[mlewis]

NovaSeq looks promising. I'm not done analyzing the data yet; for our first run read 2 had lighting failures, but read 1 was very accurate. It has other issues, though, like barcode crosstalk.

In my testing, Illumina's two highest-quality platforms are still HiSeq 2500 and MiSeq.

Thanks for the recommendation, that's good to know.

[Brian Bushnell]

I have now partially analyzed a non-failed NovaSeq run. The quality is almost as good as HS2500, but not quite. So it's extremely good. Also, the quality drops much less for each additional cycle, and much less for read 2 compared to read 1, compared to HiSeq. Also, when the run does not fail, the quality scores are fairly accurate.

The duplication rates are much higher than HiSeq (4% versus almost zero), though, and the cross-contamination rates (>1% for single-indexed reads, subject to further analysis) are much higher as well.

It seems like NovaSeq may be interesting for long reads (>150bp) since the quality drops much less per cycle than HiSeq 2500. I would extrapolate that if allowed, 300bp reads from NovaSeq would be very useful, moreso than MiSeq.

[nucacidhunter]

Thanks for reporting your analysis results.

For longer reads we have to wait for Illumina to offer a sequencing kit as the current sequencing kit cartridges cannot be toped up to do more than 300 cycles combined R1, R2 reads.

[ianakiev]

Quality Q

Also, the quality drops much less for each additional cycle, and much less for read 2 compared to read 1, compared to HiSeq. Also, when the run does not fail, the quality scores are fairly accurate.

The duplication rates are much higher than HiSeq (4% versus almost zero), though, and the cross-contamination rates (>1% for single-indexed reads, subject to further analysis) are much higher as well.

It seems like NovaSeq may be interesting for long reads (>150bp) since the quality drops much less per cycle than HiSeq 2500. I would extrapolate that if allowed, 300bp reads from NovaSeq would be very useful, moreso than MiSeq.

I believe Illumina came up with "binning" of Q scores in the newest RTA so it "looks" better than actually is. No experience though with this system, so correct me if I am wrong...