Our new HiSeq, 1st impressions...

[james hadfield]

Well it finally arrived! One of only 23 so far on the googlemap, excluding those at BGI of course.

Our HiSeq has just completed it’s validation run and passed with flying colours. I have attached a couple of status pages for you to take a look at, the interface to this kind of data is much nicer than before and there is an offline tool we can use so it can be reviewed anytime.

The validation run was a pair of PE flowcells with PhiX at 36bp read lengths. We obtained 51.5 and 56.0Gb of data from the flowcells which extrapolates to just over 300Gb from the run at PE100bp. I was pleasantly surprised to get such a hike over the 200Gb that the instrument is sold as achieving from the first run. I hope our next flowcells are anywhere near as good. The 50+Gb was for >Q30 data and we actually got just over 60Gb from one flowcell if you take the complete yield into account.

There were some spatial effects noticeable on both flowcells with the top surface appearing marginally worse than the bottom and a gradient visible on both but more pronounced on A than B. Something to watch out for later on but the volume of quality data makes this more a point of interest than anything to worry about.

The speed of data generation combined with two flowcells per instrument is going to make the HiSeq an indispensible part of daily research in many places. A comparison with SOLiD4 and the data volumes Life Technologies are talking about shows that we have entered another new era in Genomics research, one where even the smallest of labs can realistically sequence whole Human genomes without any external collaboration.

Unfortunately one thing that has not changed is the problem of dealing with masses of data. This is something where collaboration is a real boon and can make or break a project. Few labs in the world have the ability to generate samples and analyse them to completion. Tools are becoming easier to access, Galaxy and many commercial offerings as well but this is obviously the area where we will see a lot more development.

What next...
So what is going to come out of this instrument and others like it? Is the Wheat genome going to have a comprehensive draft by the end of 2010 for instance?

And where is the HiSeq 2000 going to go? Well I am hoping the clue is in the name and Terabases are possible. If Illumina can get to 200bp PE reads and 2billion clusters per run , neither of which sound impossible to me we would be touching 1TB. It is going to take a bit more work to get to HiSeq2Tb though.

For us our first Human genome on HiSeq has started as a PE100bp run and I am hoping we get over 100Gb on this run as well. I will post an update once it finishes.

[brasj]

Hi James,

just an FYI - we have been running our HiSeq for a few weeks now. When we tried to push up the clusters (one run only), we got >80Gb on a 50x50 run. So it looks to me like it should not be too hard to get a 100x100 run close to 200Gb.

Anyway, exciting times!

[KorNor]

James,

Congrats!
Hope you have fun playing with your gear.
We've had the chance here to run a few flow cells with 2 x100, as well as multiplexing on the HiSeq. So far it seems pretty user friendly, and we've had some decent results on samples as well (>100Gb).

We have taken up spiking in 1% PhiX into the samples, and ensuring that we use at least one un-biased sample in the flowcell. This has allowed us to drop the PhiX control lane, and we haven't seen much 'bleed-over' of the PhiX genome into the test library.

Like you, we have seen spatial effects on the flow cells. Usually it is low-grade, with lighter edges on both surfaces, especially on the lanes 1 & 8.
However, we did have one poor run - the flow cell functioned so poorly that focussing did not happen correctly. Multiple lanes were affected. See thumbnails from lane 7 - one whole swath is lost, and the middle swath is in very poor shape. We did have this run completely replaced by Illumina.





This is something to look out for - we check thumbnails after reviewing the first base report, and do periodic real-time checks.

cheers and good luck!
K

[james hadfield]

Hi KorNor,

We always try on the GA to have a DNA sample in lane 4 for focussing and as the control lane. I suspect we will end up doing the same on HiSeq where possible.

We are spiking PhiX in at about 0.5% in lanes 1-7 and 2% to lane 8. This shows us we have put the flowcell into the instrument and cBot in the correct orientation. Illumina failed to give us flowcells that cannot be oriented incorrectly and when sequencing very similar genomes it can be difficult to sort out any issues without this.

We also run a multi-genome alignment now so we can see the PhiX level after the pipeline has finished. This also tells us if ther is contamination from other genomes in the sample as another QC.

[mgogol]

We also run a multi-genome alignment now

I like this idea.

[NextGenSeq]

Is that a 4% error rate by cycle 40? If so, what is the error rate at 100 or 150?

[mgogol]

This thread is now a story here, kind of funny. http://www.genomeweb.com//node/94688...q_v=70aea0f8fb

[randy10]

the price tag on this is steep. Our facility is pressured by budget issues. Do you really think it is worth it?

[csquared]

James,

Your error rates appear a bit high, especially flowcell A. Also, the % >Q30 is lower than we see for a 36bp run. We have been seeing 90-93% >Q30 at cycle 50. If your experience is similar to ours, you may see lower qualities than expected as you get out to 100nt runs as we have seen quality drop off beyond cycle 50.

The gradient on the flowcells where the center of the lane has brighter/more dense clusters than the edges is a known problem that Illumina is slowly but steadily improving on.

The HiSeq is far from perfect but it is worth the investment. Our paired-100 flowcells have yielded a low of 142G and a high of 180G per flowcell......we were surprised too.

[cella]

Hi James,
Thanks for sharing.
How about the prephasing value in your HiSeq?

[james hadfield]

I will post an update as our next few runs finish. We have a PE100 on now and two more to go on as soon as it is complete, then some RNA-Seq. I will try to remeber to pur the phasing info on the screenshot.
So far run two is not quite as good with a spatial effect our FAS could not immediately explain. I think we may lose a few Gbp of data which compared to our runs a year ago seems terrible but I think is something we will all need to get used to.
"sorry you only got 148Gb" does not sound bad, but if you were expecting 200Gb it is not good.

[Lee Sam]

We also run a multi-genome alignment now so we can see the PhiX level after the pipeline has finished. This also tells us if ther is contamination from other genomes in the sample as another QC.

Hi James, we just got our HiSeq running samples and I'm wondering what genomes you run against?

[james hadfield]

I will start a new thread on this as I think it is a really useful QC for any high-throughput sequencing system or lab.

Currently we run against: Human, Mouse, Rat, Xenopus, Arabidopsis, C.elegans, Yeast, Bacteria and Viruses (the last two being amalgamations of >1500 genomes each).

A sample of seqeunce reads are taken from a lane and 36bp are aligned using Bowtie.

Reads can align to multiple genomes (conserved regions) and in this case reads are assigned to the genome with most reads. This approach should show up cases of genome contamination.

We also use Exonerate to identify sequences containing Illumina adapters.

The output is a descending list of genomes with the highest number of aligned reads first. Hopefully the genome the user was expecting!

[mendel]

HiSeq vs GAIIx

Hi
Is any have numbers for sequencing cost between Hiseq vs GAIIx?
This is for RNA-Seq experiments. Also is it possible to run 1x36 on HiSeq like GAIIx?

Mendel

[james hadfield]

Illumina don't currently sell a kit for this but anyone who would like it should make a request through TechSupport or their sales rep. Pressure of requests is likely to make it happen.

James.

PS: Anyone at AGBT who bumps into an Illumina rep could mention this so they get bombarded wth the message ;-)