Tuesday, December 10, 2019

Is ONT Really Going to Twist the Watch Stem Backwards?

I was in New York last week for the Nanopore Community Meeting.  I'll save some technical updates for tomorrow, for I wanted to focus today on a huge inflection point that Oxford Nanopore announced that they will really will punch through next year.  I'm going to do that with an extended comparison to Robert Bloch's wonderful short story That Hellbound Train.  I won't spoil that story, but if you want to first go find a copy to read (there is apparently now a graphic novel version) or spend a half hour on a decent narration of it, please do.  It's especially a joy if you're steeped in the lore of American railroading.

In That Hellbound Train, a young hobo named Martin of little ambition and significant moral flexibility is looking to hop a freight train when instead there appears a passenger train with truly frightful brakes and a strong aroma of brimstone.  The conductor seems both familiar yet strange, his hat pushed up as though something more than hair was underneath it.  Definitely not my usual ride home of the 5:35! 

The conductor offers Martin a deal -- again, never happened to me on the commute home.  Martin may have any wish he desires, so long as after his life ends he agrees to ride the train to its final destination.  Martin asks for the ability to permanently stop time so he can live a perfect moment forever, so the conductor gives Martin his elaborate gold pocketwatch with the proviso that Martin will be able to stop time precisely once.  The mechanism for stopping time is simple: just twist the watch stem backwards until the timepiece halts.

For someone like Martin the conductor's watch represents the ultimate dream, the ability at any moment to forever lock in the good times.  But therein lies a catch one might term infernal: it also means locking out any possibility of finding an even better moment.  Accepting a good now means forgoing a spectacular later. Novelty-seeking forever staunched.

Oxford Nanopore has always been a spectacular innovator, bringing forth a technology whose now near-ubiquity can camouflagege its absurd unreality.  Soap bubbles shipped around the world.  Hundreds of single molecule measurements made in parallel in a handheld device, with no need for air tables, granite slabs or Faraday cages.  Gigabases of sequence data in under an hour.  And so on.

Those of us who have been on the ONT train for the past five-and-a-half years haven't had an easy ride.  ONT has routinely been fixing bugs and driving the overall performance upwards and onwards.  The truly barely useable reads of 2014 that came from unreliable flowcells that on a good day yielded half a megabase in two days  have been replaced with raw accuracies in the mid-90s and consensus accuracies, if you don't have some of the systematic trouble sequences, exceeding 99.9%.

But that pace of innovation can be a bit exhausting.  Just this year Oxford has gone through three revisions of their R10 chemistry (pre-release, the initially released chemistry finally dubbed R10.0 and an R10.3 chemistry to be released imminently).  The FAST5 data format has gone thru many changes; I'm not surprised that the Julia code I wrote in 2014 to read it is broken but to find that tools published this year can't deal with the latest format is painful.  Plus there's now multi-FAST5, which makes managing datasets much easier by cutting down the number of files, but many tools haven't caught up to yet -- even many of ONT's own research-grade tools.

A sizable fraction of the research community has been willing to buy into this devilish bargain, that the only stable expectation is instability and chaos.  Toolchains that require constant revision and frequently break.  It also doesn't help that ONT is half-submerged their own technical turbulence: in addition to the research tools not reading multi-FAST5 there's a complete lack of consistency in how basecaller models are named across their different tools and flowcell chemistry isn't reliably recorded in the FAST5 files.

What is tolerable for a slice of the research community is intolerable for the entirety of those developing production assays for areas such as the clinic and forensics as well as many others.  If you are putting a diagnostic system in thousands of community hospitals across the country, you are sunk if at irregular intervals you must tell all your customers to patch their code.  Nobody wants to keep re-calibrating their assay, even if the newer reagents work somewhat better than the older ones.

To address this CEO Gordon Sanghera announced the Q Line, which will commence with GridION in January of next year.  Q presumably means quality, though it also hearkens to Desmond Lllewelyn.  Unlike any of the devices James Bond deployed in the field, Q Line instruments will carry some ISO registrations.  Implied is a long-term goal of appropriate regulatory approvals for use in diagnostic devices.  Sanghera touted that the management team for Q Line consists of highly experienced individuals, combining for over a century of history in regulated diagnostics (one of those statistics that if you're caught in them are both good and painful for the ego, since it simultaneously highlights your skill and age).  

But now the tensions can begin.  Sanghera believes that Q Line will represent an appropriate division of the company into an innovative, rule-breaking group and a more stable division focused on end markets.  But there will always be the strain of deciding what should move on to Q Line and what shouldn't.  The initial roadmap displayed by Sanghera has GridION and MinION Mk1C joining the Q-ranks in 2020 along with R9.4.1 and R10.3 flowcells as well as R9.4.1 in Flongle format, with the LSK9 (called just Kit9), 16S, native barcoding and rapid PCR barcoding available as library kits.  Software will initially be MinKNOW 19.11 with later upgrade to MinKNOW 20.06 and somewhere EPI2ME.  The layout of his slide leaves ambiguous the schedule for these launches.  2021  would see these joined by  PromethION and VolTRAX for hardware, R9.4.1 on PromethION and R10.3 on Flongle for flowcells, and the 96 barcoding, rapid barcoding and unspecified VolTRAX chemistries and only a single MinKNOW upgrade.

But there will always be customers pushing for greater breadth of products in Q.  Note, for example, that the rapid field kit is not on that manifest.  Nor any kind of Cas9-based enrichment.  Nor R10.3 on PromethION.  Plus there will be customer pressure to enlarge the quality-of-service provided by Q Line -- more levels of documentation, access to complete lots of kits and so forth.  

Unlike Bloch's protagonist Martin, ONT gets multiple stopwatches (or is it multiple twists?).  New products made available to the research groups will at some frequency pop up in the Q Line several years later.  But it can't be overstated what a huge cultural shift this represents, for a company that just announced its fourth (or is it fifth) basecaller architecture.  Q Line represents something close to a technology transfer operation to an entirely different company, which is quite an undertaking.  Run correctly, Q Line will become ONT's toughest, most persnickety customer. demanding excruciating levels of detail on processes and materials, being far more needling than even the whiniest external bloviator.  

The prize of course are big, stable markets for diagnostics and agriculture, markets successfully colonized by winners (Illumina) and survivors (Ion Torrent) in sequencing but not by a fleet of also-rans (SOLiD, QIAGEN, Roche 454) nor arch competitor Pacific Biosciences.  Once successfully seeded, the reticence of these spaces to innovate becomes a commercial virtue, providing an inherent barrier-to-entry for any upstart competitors.  

Can ONT pull this off?  Can they really strike the pose of Janus with an innovative side and a stable side?  Trying to run under a single roof both such operations has a very rocky history, with AT&T failing to capitalize many Bell Labs innovations being one example (though U.S. antitrust law played a significant role there) and Xerox failing to commercialize many of PARC's results another.  The two divisions of ONT will often have goals at cross-purposes, and in particular will often have conflicting demands on ONT's capital resources and managerial focus.  But it's not an impossible remit, just a very challenging one.  Only time will tell how well ONT's management rises to the challenge.

1 comment:

James@cancer said...

Hi Keith, thanks for the post.

The Q line is a must and having a restriction on product availability is better than not having any products at all! The ONT technology has promised so much for the clinical community - from day one in 2012 (http://enseqlopedia.com/2012/02/oxford-nanopore-confirmed-at-agbt) I've had a vision for rapid PCR cancer diagnostics.

It was great to see Cyclomics present at LC'19 but they'll need the Q-line to get into the regulated environment.

I know I'll watch as the research grade pulls fast ahead with respect to yield and quality and wish I was on that train. but my slow commuter line will do me fine!