Sunday, February 25, 2018

PromethION: Straining at the Starting Gate

Due to the usual time conflicts, I've only watched bits-and-pieces of the Winter Olympics from South Korea. Which is unfortunate, as I do enjoy observing many of these events as so many combine grace, power and finesse.  In the various timed events, the competitors can be seen tightly wound, ready to spring out at the crack of the start.  Increasingly, that is how Oxford Nanopore's PromethION looks: a superb performer ready to bolt away.
PromethION has had a long road and naysayers such as myself have in the past suggested ONT made a huge mistake investing in it.  Only a small number of outside labs have gotten the boxes -- my rough count is there may be as few as 5 6 installed (York, VIB in Belgium, KeyGene in Holland, Toronto and Brisbane  Melbourne [& U Oxford]).  But it is clear that the instrument is a thoroughbred capable of great things.

First, on the last day of AGBT Ken McGrath gave some overview of his experience with the machine.  Unsurprisingly, the initial flowcells were erratic in performance.  But McGrath has seen marked improvement over three shipments in both total yield and the consistency between flowcells.

But perhaps more so, McGrath's group in Brisbane, VIB and KeyGene have been posting better-and-better yields on Twitter, with the record changing weekly or almost daily.  VIB hit 63 gigabases on February 15th using human DNA, McGrath in his talk on the 15th spoke of an ongoing run that looked like it would go north of there  and KeyGene reported on the 20th hitting nearly 76 gigabases on a lettuce genome run -- only to be topped on the 21st by VIB with an 85 gigabase run!  

In the meantime, Clive Brown is claiming that ONT is routinely hitting in excess of 150Gb per flowcell.  ONT has always done better than field sites, so those numbers probably won't be seen for a while, but Brown's tweet did show a substantial boost from "active unblock", which apparently wasn't part of any of the field runs.  This is a software scheme which detects pores that have become blocked and reverses voltage on them to clear the pore.

From a scientific standpoint, PromethION could be a real beast, enabling de novo long read assembly to become standard for both known genomes such as human and complex eukaryotes.  McGrath gave a taste of the power for metagenomes: on a mock metagenome 1 minute of PromethION data was sufficient to enumerate the species composition of the mixture.

Even at the maximum price of $2K per flowcell and 50 gigabases output, PromethION would be delivering an awful lot of data for a reasonable price; at $625 a flowcell and 90 gigabases it is making long reads crazy cheap.

As Oxford ramps up, there will probably be speedbumps.  I've now spoken to a number of GridION owners, and at least two of them received dud machines -- and in one case the replacement failed soon after arrival.  But ONT has a big enough stock they can simply ship out new instruments.  PromethION will presumably be on a similar model.  Flow cell production scale-up could also be glitchy, but since Oxford has succeeded in delivering to Australia suggests they have really mastered worldwide delivery of their consumables.

Perhaps the prime thing that will prevent PromethION from completely taking over this year's London Calling agenda is that so few labs have one right now.  But with more machines anticipated to arrive at some key labs in the next few months, who knows?  Perhaps half a dozen or more of the plenary talks will detail experiences with PromethION -- and detail its use for genome assembly, metagenomics and more.  And there doesn't seem to be any serious whale watching on PromethION -- imagine the number of monster reads that might be netted

Will PromethION push Illumina to respond on price?  After all, it is reputed (particularly from Illumina investor presentations) that Big I has huge margins on reagents.  But people I spoke with AGBT who aren't committed MinION users remain skeptical of ONT data  Illumina is also projecting that their revenues from clinical-facing applications may surpass research applications this year.  So if Illumina doesn't feel heat in the clinic, they may not feel any pressure to rein in pricing.  Most likely, it could be a year or two before Oxford -- even if they execute perfectly -- can really start to peel substantial numbers of users away from the familiar reliability of Illumina.

PromethION presents an even tougher problem for Pacific Biosciences, given the leap in output per dollar the new box represents.  PacBio retains a significant advantage in consensus error quality, but as I've noted before, many applications aren't very sensitive to this.  Full length RNA sequencing is one example; if you have a good reference genome getting the mRNA sequences base perfect isn't critical.  Again, ONT has a hill to climb in terms of community acceptance and perception of the data quality -- but now PacBio must fight a two-front war of arguing data quality over raw cost.  Niches such as large-scale synthetic biology clone validation play well to PacBio right now, but those haven't proven to be large markets.  And PacBio isn't deeply entrenched in the clinic, so they don't have clinical markets as a refuge. There's no doubt PacBio is a sharp organization filled with clever individuals, but they are now faced with a very difficult competitor.

All of this could get even messier for the competition in a hurry.  Once Oxford is fully ramped on shipping PromethIONs -- or given that this is ONT, perhaps well before this -- expect a version of GridION that takes PromethION flowcells instead of MinION ones.  Clive Brown has hinted at this, and it would certainly have a market.  Ideally such a box would enable quick swapping from one mode to the other, but I suspect that instead such boxes will come either all MinION or all PromethION.  Such a box, with a lower initial price point and taking less bench real estate, could appeal to smaller labs.

As always -- stay tuned! Luckily we don't have to wait four years for the next exciting developments in sequencing technology.

[2018-02-26 8:10 corrected Brisbane to Melbourne and added U Oxford, which I had missed]


svennd said...

Active unblock was released on 15.02 for promethion, and I believe both for keygene and VIB/ua the active unblock updates where installed, so don't think we will see huge increases soon. Without active unblock we reached 63Gb ( with it 85GB

They also where both ran on friday 16.02. However the public reporting happened a bit later since its difficult to accurately estimate the yield during run w/o live basecalling. (guppy should be a game changer soon)

Dale Yuzuki said...

Thanks for this Keith. For clarification, is the $625/flowcell price something that was floated as a ‘at high volume’, and the $2K number a current price?

Anonymous said...

FWIW Melbourne, not Brisbane

Keith Robison said...

Dale: I should have given the link -- to get the $625/flowcell price one must commit to buying 720 packs (2880 total flowcells) for $1.8M
Nanopore Flowcell Pricing

Jeff B. said...

Throwing out throughput numbers is meaningless. Not all bases are created equally.

A 30x human genome with ONT only gives you ~QV20, and ~QV30 will illumina corrected reads.

You can get to QV30 with ~10-15x coverage with Pacbio.

A 30x ONT genome will not be anywhere near the equivalent of a 30X Pacbio or Illumina genome.

Mike said...

Yawn. This product is still far from being able to compete with illumina or even Pacbio. A few beta sites that sequence 1/cell per week doesn't really impress me. ONT hasnt proven they can reliably scale a product. The GridIon was a complete failure. Does anyone know a lab that uses the GridIon at anywhere near capacity? Has any lab had repeat orders for the GridIon? Just take a look at some quotes from a recent research report about ONT
An employee of one of the three Chinese buyers Nanopore recently announced
confirmed that the company had purchased three sequencers from
Oxford Nanopore, but “We’re not very satisfied with the results. We definitely
will not buy more,” he said. “Their technology is not very mature….
We spent all this money, but we achieved about 10% of the expected results.”
He added, “We think about 80% of what company management says
is bull.”

"A genetics institute in Beijing conducted a detailed comparison of existing
technologies for long sequencing. One of the researchers said that
Nanopore’s reads are the longest but that the institute chose to buy PacBio
equipment. A management employee of a key partner to Nanopore in a
phone interview said [our translation], “The cost for Nanopore reads is very
high; basically it’s very hard to make commercial. Even though the machine
isn’t that expensive, the materials are, because you need to do a lot
of tests.” "

Meanwhile Pacbio secured multiple repeat orders last year from NovoGene, GrandOmics, BGI, Annoroad, and Histogenetics.

All I see here in ONT rushing out a product that works when you hand select your best flow cells to a few customers to ensure maximum hype. I guarantee the promethion won't be used for any large scale projects this year. ONT will have major growing pains. They're just rushing out data and making twitter claims of "cheapest" cost per base because they are in desperate need of a funding round.

Anonymous said...

"A 30x ONT genome will not be anywhere near the equivalent of a 30X Pacbio or Illumina genome."
Why don't ONT just sequence the product of a CCS style substrate? It's not like it is a new concept, people did it with Sanger, so I don't see why not?

Duarte Molha said...

I wish I could understand what the last comment is talking about. Just when I think I am getting the hang of my topic someone just reminds me of my ignorance :)

Keith Robison said...

Anonymous: There is a protocol for this, using rolling circle amplification to generate a concatenated series generated from the same template. ONT can't do a circular read like PacBio, since the molecule must be threaded through the pore as a single strand and then can't return.

But that doesn't address the fundamental problem that at the level of signal different polynucleotide runs look the same -- so no matter how deeply you sequence the same template, there will be ambiguity/error