Opiniomics

bioinformatics, genomes, biology etc. "I don't mean to sound angry and cynical, but I am, so that's how it comes across"

7 reasons why the NextSeq 500 is a strange choice

Illumina have released two new sequencing instruments recently, including the NextSeq 500, which I think is aimed at killing the Ion Proton.  I have to admit I am a little confused by the NextSeq 500 and I wonder what niche it fills?  Here’s why:

1. The HiSeq 2500 is much higher throughput

Many are billing the NextSeq 500 as a genome-scale machine in a benchtop size box.  However, let’s look at the stats .  The best output you can get from a NextSeq 500 is 120Gb, which is one human genome at 40X coverage.  But who wants to do one human genome?  By my calculations, 120Gb gets you about 35 exomes (assuming a 60Mb capture @ 40x and 70% on target).  Again, who is doing 35 exomes?  To find causal links to disease, you need to be doing 1000s of exomes.  Illumina state that the NextSeq 500 is for “everyday” exomes.  I assume this means the clinical market, but I remain to be convinced this is the best route.

Contrast this with the high output mode of the HiSeq 2500, recently given the 1Tb upgrade.  That’s 1000Gb, or 8 times the output of the NextSeq.  That’s 8 human genomes, or ~290 exomes.  Clearly the HiSeq 2500 is better for medium- to large- scale projects.

Then there is time – the NextSeq takes 30 hours to get to 120Gb and the HiSeq 2500 takes 6 days to get to 1000Gb.  So in theory you can run the NextSeq five times in the same time it takes a HiSeq 2500 to run (though you will be in the lab at some very strange times) – but this still only gets you 600Gb compared to the 1000Gb the 2500 is giving you.

There is no real difference in costs for the sequence data, so I’m left asking myself – why would you buy a NextSeq 500 when you can buy data from someone with a HiSeq 2500?

2. The HiSeq 2500 is more flexible

The NextSeq 500 has one flowcell with four lanes, whether run in rapid or high-output mode.  The HiSeq 2500 has 2 flowcells of 2 lanes in rapid run, and 2 flowcells of 8 lanes in high-output, and each flowcell can be run independently.  This means you can configure the HiSeq 2500 to run 2 lanes, 4 lanes, 8 lanes or 16 lanes.  Given you will multiplex multiple samples in a lane for most experiments, the HiSeq 2500 clearly provides far more flexibility.

3. The HiSeq 2500 has longer reads

OK, at the moment both systems do 2x150bp, but 2x250bp is due for release on the HiSeq 2500 later this year.  It’s easy to speculate that the same will come to the NextSeq, but I wouldn’t be so sure.  Data shown at PAG demonstrated that NextSeq data is already of a lower quality than HiSeq 2500 data, and longer reads usually mean even lower quality.  So for now, the HiSeq 2500 is the choice for high-throughput long reads

4. The MiSeq is cheaper

If you really want to get into NGS then the MiSeq has just dropped to between £60-70k.  This little beauty will give you 25 million reads at a variety of lengths and excellent quality.  It’s not as high-throughput as the NextSeq 500, but as I’ve discussed in point 1 above, the NextSeq 500 isn’t really high throughput.  Perhaps the best option is for labs to buy a MiSeq for small-scale and then outsource their medium- to large- scale projects.

5. The MiSeq has much longer reads

The MiSeq is a beautiful machine and will deliver 2x300bp reads of amazing quality.  We’ve been promised 2x400bp reads for some time, and I personally cannot wait.  For long reads of Illumina quality, the MiSeq is the weapon of choice.

6. The NextSeq 500 has a new two-dye system

The NextSeq 500 has a new two-dye system for measuring base-incorporation, with red dye measuring C, green dye measuring T, a mix of dyes measuring A and no dye measuring G.  I think it’s the latter two that are causing most concern.  I’ve little doubt that this new system will introduce bias.  Sure, it will probably be solved by chemistry and software upgrades, but when?

two-dye

7. You don’t have to spend $250k

Firstly, the cost of a NextSeq isn’t $250k, because you will need a library technician, a bioinformatician, some compute and some data storage, so start thinking more in the “$500k start-up and $250k per year” ball-park.  I am sure there are better things you can think of to spend your money on!

Secondly, you don’t have to spend $250k because you can already buy Illumina sequence data, and there are many reasons why you should out-source this particular part of your scientific research.

So all in all, I am confused by the NextSeq 500.  As I have said before, I think it is designed to compete with the Ion Proton, but looking at the statistics, I am not sure which niche in the market it really fills.  Comments welcome if you care to educate me – is anyone planning to buy a NextSeq 500?  I’d love to hear from you!

 

47 Comments

  1. Illumina has yet to say how they will upgrade any machines purchased before 2013 so many people may not even gain access to the 2×250-bp run length. Another question is what will the quality look like for a run that long using either chemistry scheme?

    • Illumina said that it is firmware and software only; no hardware updates required. This was during one of their NextSeq 500 seminars.

      • Firmware and software only on the new Hi-seq upgrade, you are correct. The reason why it is only available on the newer Hi-seq’s (2013+) is because these ones had the new camera/optics on them. I was digging into these details myself with Illumina.

  2. What is the per run cost? In other words, how much do the SBS reagents cost per run on the NextSeq 500?

  3. So the short answer is that a research institution with middle-tier funding and multiple model organisms would like a way to sequence complete eukaryotic genomes and transcriptomes at a reasonable price. The Core Facility at this institution would like to provide rapid turn-around, cost-effective sequencing to the researchers. A Nextseq looks like a reasonable entry-level sequencing system for these needs.

    • But those institutions can buy sequencing services from any number of providers for a far cheaper (real terms) cost.

      • Geneticist from the East

        23rd January 2014 at 5:00 am

        I think for some clinical application (e.g. cancer) that turnaround time is critical. Then buying NextSeq 500 can make more sense than outsourcing the job.

        • As I’ve said previously, the time limiting step for a clinical decision is not going to be the sequencing – it’s going to be the bioinformatics followed by the clinical interpretation of the results. Bear in mind that in the UK it takes a few weeks (for cancer; several months for other diseases) to even get to see a consultant. In that context, the time it takes to generate sequence is irrelevant.

  4. We at AllSeq certainly wouldn’t argue against the wisdom of outsourcing your sequencing needs. Unless you’re willing/able to ‘feed the beast’ 24×7, it can be much more economical to outsource. If, however, you’ve made the decision to buy your own instrument, I think you might have a hard time justifying the purchase of a single HiSeq 2500 over the purchase of three NextSeq 500’s. For the same capital outlay you’ll get a set of instruments capable of generating ~300Gb per day (nearly double that of the 2500), roughly the same operating costs ($33/Gb vs $29/Gb) and increased flexibility (three flowcells vs two). You can compare the specs of Illumina’s systems on our NGS Knowledge Bank: http://allseq.com/knowledgebank/sequencing-platforms/illumina/

    Of course, the biggest unknown for the NextSeq 500 is what that 2-dye data is going to look like. Can’t wait to find out!

    • You can get a 2500 for a lot less than $750k, I’m unsure where that figure comes from. And the 2500 is cheaper per Gb, and more flexible. I still don’t see the NextSeq market.

      • Can you really get a HiSeq 2500 for a lot cheaper than $750k? I’m pretty sure the list price is $740k. Of course there is the possibility of discounts, but probably not too much when buying a single instrument. And any discounting would probably apply to the NextSeq 500 as well.

        It will be interesting to see how the market responds. I’ve heard one anecdote of a customer switching an order from a 2500 to a NextSeq 500, but an N of one does not a market make. Let’s see.

        • Whether someone did it and whether makes sense are two different things 🙂

          Traditionally Illumina have a policy of not discounting new machines, so no NextSeq discount; and yes I believe any 2500 sales will be heavily discounted.

          You’re a provider: why would you buy one?

  5. So now I’m confused. You say it doesn’t make sense to buy a NextSeq 500 and now you’re saying it doesn’t make sense for a provider to buy a 2500? What’s left? A MiSeq won’t cut it for most applications and 4 of the 5 potential HiSeq X Tens for 2014 have already been sold (and very few providers can probably afford the minimum buy-in anyway). What would you expect a provider to do if they want to increase their capacity?

    • My apologies. I meant would you buy a NextSeq? And why?

      • Oh, got it – that makes more sense. I think you’re right that providers are more likely to buy a 2500 while individual researchers are more likely to buy the NextSeq. I guess Illumina will find out if there are enough individuals whose needs fall between the MiSeq and the HiSeq (regardless of whether or not they should have just outsourced instead 😉

  6. I’m looking at the NextSeq for my clinical molecular lab. We are looking to do about 10 exomes a week and this will give us the perfect throughput. The MiSeq is too small and the HiSeq is too big. We were considering the HiSeq1500, but after some serious thought we are going to pull the trigger on the NextSeq.

    • And you couldn’t outsource?

      • Since we are a clinical lab of a hospital we are CLIA licensed to run patient tests and bill insurance for our services. The only way to outsource the sequencing component of the test is to have another CLIA lab run the test. There aren’t any good options for this and the two that exist are charging way too much. Currently our patients get their testing performed at other CLIA labs that do the sequencing, analysis of data, interpretation of variants, and reporting of results. I can’t bring this testing in house without bringing all aspects into a CLIA environment. Also, since my name goes on the patient report I am liable for it. If we outsource the sequencing it would be putting a lot of faith in the lab performing this component of my clinical test.

  7. In a word, RNAseq (profiling not full transcriptomics). Local genome centres cant fill a Hiseq flowcell with 50/75PE samples and keep to a reasonable turnaround time. Slightly higher base level error not a disaster, cheaper than a HiSeq2500 rapid run (which are hard to find as Hiseqs tend to operate on full runs), and as for the other infrastructure, we have all that anyway. Same (ish) pipeline. And its reasonably affordable. Don’t forget outsourcing adds 20% VAT in the UK. With a Nextseq we would be significantly cheaper than external suppliers for small projects. And thats what we do. Lots of small projects.

  8. We’re a core facility and I think the NextSeq would suit is quite well. Much of the sequencing we do is for relatively small projects and the vast majority of our work is done on the MiSeq instrumentation for it’s small yield, fast turn-round and (relatively) long reads. Often these projects are microbial genome sequencing, microbial transcriptomics, 16S survey sequencing but we also do a lot of small eukaryote-based projects like small numbers of exomes, small RNA sequencing, whole genomes for novel organisms etc. Usually when we run HiSeq projects, the customers are buying only one or two lanes at a time. Lastly, the upfront instrument cost is a bit issue for us – we spend our own funds, not grant money, on new infrastructure. $500K difference is a big deal.

    • I guess my challenge to you is that, instead of setting up your own core facility, you could buy data (more cheaply and more reliably) from an existing next-gen provider; I am suggesting that this would be a better use of your money 🙂

      • Primarily because we’re not setting up a new core facility… we’ve been operating for about 7 years as an NGS provider (since the bad ol’ days of the original Genome Analyzer) and about 25 years in total (Sanger/CE and other technologies).

  9. EmbarassedforAuthor

    18th March 2014 at 1:31 pm

    Why would anyone choose the NextSeq? Is this a serious post?! Perhaps, because you have to purchase 10 HiSeqs, whereas option to purchase 1 NextSeq exists. Surprised I had to state the obvious. I can’t believe I even read this whole thing.

    • I would have liked it if you *had* read it.

      Null hypothesis: it is cheaper and more reliable for you to buy data from an existing provider of next-gen services than it is for you to buy and run your own NextSeq.

      Disprove the null hypothesis!

      • Operating a core facility isn’t always about cost. It’s about training, accessibility, convenience, flexibility, customisation, application development, etc. We can also provide a much better turn-around than many providers.
        But, you’re a service provider, as am I, so I guess it’s in both our interests to discourage people from buying these “cheap desktop” sequencers

        • I don’t mind people buying them, I honestly don’t, I’d say 90% of people I know who’ve bought a benchtop sequencer don’t use it anywhere near it’s capacity, and it seems a waste.

      • Biomickwatson,
        I think the argument to outsource is strong… especially if you are not doing hundreds of samples a year. We did a post on this below sometime back (when the mi-seq came out).
        http://cofactorgenomics.com/average-total-cost-atc-real-cost-sequencing-run/
        (looks like we need to update it with the NextSeq or Hi-Seq upgrade?)
        However, ScottC is correct. If you don’t have to invest in the infrastructure and you have more than a decade of experience sitting there waiting to take advantage of… then machine husbandry makes a lot of sense. I think most people we work with (self selective I’m sure) want the benefit of the technology for their research, but don’t want to have to learn about the intricacies of balancing multi-plexed pools of samples, modifying and testing protocols to limit ribosomal content, and how to write software to distinguish between biological noise and noise introduced in the sample prep.
        With that said, we work with a number of academic institutions that have their own sequencers and have the machines because of department clout of having sequencing capability, being able to educate students to the technologies, etc. However, when it comes time to do challenging projects or precious samples are involved, many times the head of the dept. will recommend they get outsourced to specialists in a given application.
        I personally think the NextSeq has a place (the reason why we have added it to our roster of 5 platforms we support recently). 75% of the work we do is RNA-seq, and it proves nice to have the option to generate the equivalent of 2.5 lanes of a High-Seq in a day. The combination of price/turnaround/capicity has a fit in our group (it may not make sense for all groups). We estimate this platform will run about 1600 RNA-seq samples from Cofactor in the next year.

      • It is never cheaper to buy data. Most sequencing centre in Canada anyway are charging in the $1500 to $2000 range per sample. Frequently is it also not faster. It has taken up to 6 months from time of submission to get data back. By taking on the equipment (equipment grants here are separate and cannot be used for anything else, so if they are giving out the money why not take it) and handling of the samples (which is largely kit driven) the cost will come down to $3-400 per sample or even less depending on the need read depth and multiplexing. Clearly the sequencing centres have overheads, labour costs, but so do I. I am already paying the students and techs, why shouldn’t they become more familiar with this technology? I agree a large clinical project would be better suited on a bigger instrument, but when we have 8-10 RNA-seq samples or 5-6 ChIPs this is the right size instrument.
        I agree there are head aches with starting new technology. I used to run a proteomics core. But these kinds of technological improvements are meant to bring sequencing to more labs. Additionally more instruments will drive innovation. When a research group has an instrument it will be used in different ways than you would predict.

        • OK, firstly, I think you’re with the wrong supplier. We can turn around some samples in a week.

          Secondly, your costs: $1500 – $2000 per sample – of what? Our Illumina sequencing costs range from £12 to £2500 per sample, depending on what you’re doing.

          Thirdly, this is the most misunderstood point, and everyone who I speak to completely misses this point. Let’s say your post-docs and students are doing something now that is not sequencing. If you stop them doing that thing, and make them do the sequencing, then the cost of the students and post-docs is now a cost of your sequencing. Why? Because the thing they were doing before is now not getting done. This is what’s known as a hidden cost and some really very intelligent people I chatted to on Twitter completely do not get this point. It’s like economics 101, high school stuff that people don’t get. It confuses me why!

          Fourthly, I don’t understand why people don’t just admit – the real reason is they want their own kit, because it’s cool kit and they want to own it.

          Fifthly, not a single one of you are factoring in the amortisation of the machine. In 2 years, you’ll need to buy another one. Or upgrade it. How are you going to do that?

          Sixthly, what’s your plan for maintenance? You got a contract? What happens when the NextSeq breaks and you don’t have a replacement?

          Seventhly, you figure out how to deal with that wash through yet? (repeat this 100s of times for small niggly issues that Illumina facilities know about and people in research labs don’t).

          Anyway – *sigh* – I’m done with these arguments.

          I want everyone in the World to buy a NextSeq 500. Everyone. I want to watch them gather dust.

  10. Im in the market for a NGS machine and am considering the MiSeq and NextSeq. We are currently outsourcing out sequencing, about 300 samples/year. Running 24 -48 samples/lane on a HiSeq 2500 gives us a useable amount of data. A NextSeq would handle 10- 20 equivalent samples and a MiSeq 1-2. Trying to push 300 sample through a MiSeq sounds like a chore, wherease for a NextSeq this wouldn’t be a problem. I suppose I could go for to MiSeqs which would be less than 1 NextSeq, but I don’t know if that makes sense either. Right now the NextSeq seems to be the more logical choice.

    • But why is it better than continuing to out source? You have a huge capital cost and a staff cost, not to mention needing to learn new skills you may not have. Why not let someone else take care of that?

  11. Genohub Support

    22nd April 2014 at 10:20 pm

    If you’re keen on getting Nextseq data (data in 3-4 days) you can order immediately through Genohub: https://genohub.com/shop-by-next-gen-sequencing-technology/#query=9eed279e5558472c0b711e4cefafe42d
    @Sean Hofherr You can send a message to the provider to discuss your clinical application. The provider is CLIA certified.

    • The thing all the data generating companies do not realize is that for me to put my name on a report based on someone else’s data, regardless of CLIA or not, is suicidal. The benefit of doing it in house is that I contrall all aspects of the data generation and I know all details about how it was performed and what QC measures are in place.

      • I suspect the vast majority of publicly available sequence data was generated by “other people”. You think all of that was basically “suicidal”?

        • I take it you aren’t an ABMG diplomat. Research versus clinical is totally different. The liability involved in fee for service diagnostics is very much higher than generating sequencing data for an NIH funded study on annonymized subjects or controls. I do not know a single clinical lab that is outsourcing their sequencing.

          • Clinical testing or clinical research?

            Besides which, it’s all done by “someone else” to a certain extent- internal or external facility.

          • Clinical fee for service testing. In my hospital we send out 800 samples to about 100 different reference labs. The bill us and we charge the patient or their insurance. We get billed around $3 million a year for these labs to test our patients. If we can do a test for less than they charge us I save my institution. These are all diagnostic tests. I explored bringing the testing in house by outsourcing the raw data generation. The issue is that the final report is made by me and if I am signing it I am taking responsibility. For my other tests I know their pitfalls or if something odd is going on, but with outsourcing the raw data generation to another CLIA lab I am assuming risk.

          • You are taking the risk whether the lab is 100 miles away or in the office next door.

            My point being that the guys on the office next door will make the same mistakes the out sourced lab do; probably more because they’re not DNA sequencer experts.

          • And I assume NextEl 500 is yet to be clinically certified.

  12. Our department just purchased a NextSeq 500. The key driving factor is/was turn-around time. Our university-sponsored core averages between 5 and 7 weeks for a lane of HiSeq 50SE. The best commercial vendor times we have been quoted is 4 weeks, and that’s for an extra charge. It’s usually more like 45 days, net turnaround. There is a certain value (I believe real value, perhaps you think it’s perceived) to having your data back in a few days. To me, it’s really amazing to think I can start a ChIP-seq experiment on a Monday and have sequenced reads back in my hands on Friday morning.
    In response to the outsourcing question, the chair was much more willing to buy the instrument and pay for part of the technician than he was to giving each lab the the equivalent $25,000 in sequencing costs in lieu of buying a NextSeq. I think the NextSeq is a good recruiting tool for the department, as well, during our upcoming faculty search.

  13. We got our NextSeq in the end of July and I can now say conclusively it is not a strange choice. It is simple to use, the data quality is phenomenal and integration with BaseSpace is also great. We are running the TruSight One panel of 4800 genes and it has been performing very well with minimal optimization.

    • Sean,
      Would you be willing to share some phiX data from a good NextSeq500 run ?

      • Sure. Email me offline. My email is my last name.first name @gmail.com. If you have a BaseSpace account I can share it. Otherwise you need to tell me how you want to get it.

  14. I can’t argue with the calculation you stated-
    By my calculations, 120Gb gets you about 35 exomes (assuming a 60Mb capture @ 40x and 70% on target). Outsourcing question aside, this actually sounds like a good platform and yet Illumina say 9 exomes is recommended, why are we losing so much capacity? Not so sure about NextSeq.

  15. I have been using the NextSeq500 for 6 runs with different conditions, single-end or paired-end, high throughput or medium throughput, 76-bp runs to 301-bp runs, all with chemistry version 1. In our experiments, the reads have constant a constant part in the middle, where we observe a spectacular drop of quality. According to the machine, it is about 10-20, but in reality it is much worse. When the sequences are different again, the quality rises above 30.

    We also have sequences with a near perfect forward repeat of 150 bp, with a nucleotide in the middle that is either C or T. The machine sequences an A almost every time, so it is obviously reading both repeats at the same time (remember that a C/T mixture is the same as an A on this machine).

    We also have a whole experiment where the reads look something like this GGGGGGGGGGGGGGGGGGGGGGGGAAAAAAAAAAAAAAAAAAAAAAAAA. Unrelated with our sequence, and quite disturbing because the machine is not supposed to be able to sequence anything starting with 4G because those clusters are invisible.

    Otherwise, we can recognise the reads, we can map them but there are typically more mismatches than I was used to with the HiSeq.

  16. This does sound like a hard problem for NextSeq v1. I imagine that the mis-called middle base could be caused by phasing failure after the long repeat. The whole experiment of G(n)A(n) looks like a failure and I wouldn’t spend too much energy trying to diagnose it unless you have the Run Folder data and included a PhiX spike. Our experience is that v2 is a big improvement for the mismatches, you would likely see similar improvement, but I don’t know if the update would similarly improve the long repeat.

    There is a distinct lack of love for NextSeq online, and it was certainly marketed as a mature product when it was not. It didn’t help that Illumina pushed BaseSpace as obligatory (it is not). Our experience (mainly cancer amplicons) has been that it is easier to run than a MiSeq and the v2 chemistry makes the data quality effectively equivalent. We won’t be tempted by the next v1 product from Illumina, but I think the NextSeq will settle in fine.

    Good luck!

Leave a Reply

Your email address will not be published.

*

*

code

© 2017 Opiniomics

Theme by Anders NorenUp ↑