HiSeq 2000 sequencing report

[aleferna]

We received this report from our latest sequencing project. We've been going back and forth with the core-facility because they want us to use the data from lanes 1,2,3 but we don't want to accept the data since the error rate is so hi. Is this a normal quality report? Should we accept the data from the first 3 lanes?

We also noticed that there is a spike in the error lane in position 22,23,24 and 25, is this normal?

[aleferna]

Sequencing Report from core facility

forgot the pdf

[aleferna]

Initially since they mention in the report that 3 of the lanes failed then we thought, they will just resequence them, but now they are saying that we should join this with another failed run (this time caused by some faulty chemistry from Illumina) and that then we will get a sufficient number of reads.

What criteria can you use to accept or reject data from a HiSeq ? I know that HiSeq performance varies quite a bit, but there should be a range of what you can call a success or a fail???

[monad]

You have very noisy quality score from the beginning of sequencing. What was the materials you sequenced? Any chance this is reduced genome or amplicons? Whether passing filtered portion of data is usable or not is up to you depending on criteria you need for your data analysis, but even when you decide to use this data, that should be considered as bonus data and in general core lab should be willing to rerun the failed lanes. Just my 2cents.

[NextGenSeq]

It doesn't look that bad to me. I've seen a lot worse than this and the PhiX error rate is low. It's your library not their sequencing.

[aleferna]

But I thought these reports are only based on PhiX?? Why would my material matter? and why is the PhiX quality significantly lower in lanes 1 to 3 than in 4?

[NextGenSeq]

So you think a 0.2% difference in error rate is significant with a SD of 0.3?

The important thing is how well your data aligns to your reference genome and the quality values. I bet if they didn't say "failed" in the report you wouldn't even have noticed anything.

[aleferna]

Well, the problem is the number of reads, we were expecting much more than 8Gbp per lane. If you start with 300Gbp per HiSeq 2000 run? Is this realistic, I took the lower end of the old chem? In my head if you get 300Gbp for 2 cells each with 8 lanes shouldn't you get around 300/16 = ~18Gb per lane?? What is the real world HiSeq throughput(100bp x PE) with the current chemistry?

[NextGenSeq]

For poor libraries we'll see 80 million reads per lane or less if they are really bad. Good libraries give 100 to 130 million per lane.

Once we began quantifying our libraries by QPCR we get much more consistent cluster densities. We get good results using the NuGen kits for PE genomic libraries and the protocol is very fast and easy.

[aleferna]

When you say 80 million reads is that 8Gb per lane or 16Gb ??

[monad]

That is 80M x 200 (2x100), so 16G per lane.
Again, you need to revisit the goal of your sequencing.
Your data may have lots of room to tolerate with marginal quality or not depending on what you'd like to get out of.

[aleferna]

Thats the problem, its a very complex experiment and I usually through away 95% of the mapped reads, so I'm working with 5% of the library, therefore I get as many reads as I can.

[aleferna]

But I mean, if the core facility gives you back 8G instead of 16G how do you know if it was their problem or if it was your sample? So am I getting a crappy run because of my sample or because they still don't have a good batch?

The problem is that they've been having problems with a bad batch of reagents, they've also had problems with the machine itself and they messed up one of other samples because the light went off and they didn't have a backup, we've been waiting for this data for 6 months. So these things makes you question the overall integrity of the service. Also for some reason in both runs its always lane 4 that works the best, so I get a feeling that the machine is not 100% operational, but I don't really know whats going on in that facility.

[monad]

I guess we are talking about two different things here. Criteria for the successful sequencing from the instrument operators standpoint vs data content. Not knowing what is the nature of your input sample, it would be difficult to suggest anything much meaningful here. However, as an example, if your input is reduced representation materials, normal range of cluster density for usual situation may affect the phix control quality as they are spiked-in in your sample lanes. Chances are your sequencing data is as good as you could get. 5% of usable data has nothing to do with sequencing quality. In another word, as NextGenSeq suggested, sequencing was just fine and repeat sequencing most likely will get you same "quality" data.

[aleferna]

Have you heard any news on this "bad batch" of reagents from Illumina, it was supposed to be giving problems last December here in Europe?? Has anybody heard about this? How do you know the batch is bad?