MiSeq error rate for 600 cycle PE

[Luvs_2_sequence]

Was a bit alarmed at the run for 600 cycle chemistry we did the other day. Error profile shows that along the read, the error rate increases dramatically. Even only 200 cycles into the first read, the error rate has already surpassed 1%
The error rate for read 2 is similar.

Is this expected? We were a bit shocked to see this high of any error rate and was wondering if others were seeing that doing 600 cycle runs.

Let me know

[nickloman]

Yes, we are too.

[DJ Flowcell]

Yeah, we are experiencing the same thing. The chemistry has been improved though, although the cycle time is back up to 5min (up from 4.3min for the v2 kits). By the end of the first 300bp read, for %perfect reads, the 0 error metric is down to about 45%

[WhiteSeal]

We are seeing a lot of issues with these 600 runs. There is a significant decrease in Qscores around 200 cycles (first read) and then around cycle 450 of the second read (a picture of %>Q30 is attached). We've had a lot of data that just isn't good enough for data analysis.. and I mean a lot of data. Certainly it is not what the kit should return as a result. I mean it is a 2x300 kit. Not a 2x maybe 300 if you are lucky. We've contacted illumina about this and have received compensation for 3 kits already (hoping for more) but has anyone else tried this.

[GenoMax]

Drop-off in quality accompanies presence of short inserts (when one would start reading to the adapter on other end). Have you checked/eliminated that possibility for this data set. It seems to happen more often that one would expect.

[NextGenSeq]

IMHO the v4 chemistry isn't worth the extra money.

How many reads did you get?

[WhiteSeal]

@GenoMax: samples were checked on BioAnalyzer to check the length of the fragments (all >700bp) so this should not happen. Also, we have seen this with almost all our 600PE v3 runs on the MiSeq.

@NextGenSeq: MiSeq run 600PEv3 with 566k clusters gave - according to illumina output - ~14M reads, but in read two at cycle 550 already 5M reads have a Qscore <Q30 (not good), and at cycle 600 12M reads have a Qscore <Q30 (really not good). And then this was a 'good' run.

[nucacidhunter]

The explanation so far leaves two possibilities open. One is library diversity (particularly in the positions with low Q scores) and the other is faulty reagents. Illumine has had some recalls recently. I wonder what was %PF for the run.

[GenoMax]

@WhiteSeal: Has this data been analyzed and you are sure the sequence (irrespective of Q-scores) looks ok (no adapters etc).

Points mentioned by nucacidhunder are very relevant. If the nucleotide diversity drops late into the read then that would also lead to Q-score drops. Since you have been in touch with Illumina tech support if you used reagents that were recalled they should alert you. Wouldn't hurt to proactively check on that possibility too.

[avo]

I'm a bit confused by the %Q30 graph shown here. I have seen 600V3 runs on our MiSeq that looked much worse. Yet we generally cluster well above 566k, so that might be a factor. Nevertheless I would be perfectly satisfied to see a Q30 graph like this. Even the example runs I looked at on Illuminas BaseSpace for the 600V3 kit are worse (SmallGenomesBcereusRhodo).

Because all our runs met the Illumina specs of a Total % ≥ Q30 of 70%, i never thought about complaining.
We observe this for a variety of different samples using TruSeq LT as well as TruSeq PCR-free libraries. Based on the %Base plots and the output of the adapter trimming there shouldn't be a significant problem with the fragment size (>800 bp) either.

Despite that we lose bases from the ends of both reads during the quality trimming it doesn't have a major impact on our analysis.

Does anyone experience similar issues?

[WhiteSeal]

@nucacidhunter: indexing was fine. 76 samples, barcode variation for each base was evenly distributed. So no issue there.

@GenoMax: I have not had the opportunity to look into the data myself, but customer is complaining that Qscores at final 50 cycles is well below expected.

@avo: I was wondering.. you say you have seen worse and you just trim the data, but with so many reads not making it till the end then what is the use of running a 600PE kit. I (and customer also) expect at least 70% of all data to have a >Q30. In this case 70% is around cycle 550...

As for faulty reagents. This was actually a replacement kit from Illumina for one of their faulty ones. Also, I was proactive and complained and replacement has been received to run the pool again.

[NextGenSeq]

@GenoMax: samples were checked on BioAnalyzer to check the length of the fragments (all >700bp) so this should not happen. Also, we have seen this with almost all our 600PE v3 runs on the MiSeq.

@NextGenSeq: MiSeq run 600PEv3 with 566k clusters gave - according to illumina output - ~14M reads, but in read two at cycle 550 already 5M reads have a Qscore <Q30 (not good), and at cycle 600 12M reads have a Qscore <Q30 (really not good). And then this was a 'good' run.

This is really low. Did you spike in PhiX? If so what are those reads like?

If they are good it's your library.

[GenoMax]

@GenoMax: I have not had the opportunity to look into the data myself, but customer is complaining that Qscores at final 50 cycles is well below expected.

Should one overly worry about Q-scores, specially for a re-sequencing run (not sure what kind of a run this is)? If a reference genome was available (and you are not trying to call SNP's in that region) the sequence may be still usable (irrespective of the Q-scores). If this was a de novo run then the worry would be justifiable.

[pmiguel]

IMHO the v4 chemistry isn't worth the extra money.

40% more in reagent costs gets you roughly twice the data. Generally see >5 billion >=Q30 bases for a 500 cycle run and >10 billion >=Q30 bases for a 600 cycle run.

--
Phillip

[NextGenSeq]

On the MiSeq we get maybe 24 million versus 20 million for v3. That's not worth it for an extra $400.