2019 Tech Speculations: Oxford Nanopore

As promised in the last post, I'm segregating out Oxford Nanopore.  Admittedly I tend to cover them relatively closely -- though I never seem to quite finish writing up their conferences -- but at the moment ONT is the only major player in the U.S. research sequencing market not being run out of (or about to be run out of) Illumina HQ.  And I'll be very to the point: ONT has a lot of balls in the air and irons in the fire, but from my point-of-view what matters most is rapid and regular progress on the accuracy front.
I doubt many will argue with me that Oxford Nanopore is the competitor most affected by Illumina's ongoing (and delayed by the U.S. government shutdown) acquisition of Pacific Biosciences.  ONT and PacBio are the only players currently in the true long read space and have dueled each other in the courtroom (with ONT just winning a case in Europe).  Now instead of a small company with an uncertain future and a small salesforce, ONT must compete always with the 800 pound gorilla of the sequencing world.  I expect Illumina salespeople to aggressively cross-sell the two products.  And, based on Illumina CEO Francis deSouza's J.P. Morgan talk, Illumina will be pitching themselves as the most accurate systems.  PacBio has rebranded the Circular Consensus Sequencing as HiFi, with 20 kilobase inserts getting multiple passes and a revised software stack speeding the downstream processing.

The accuracy of the Oxford Nanopore platform has always been an issue and has undergone a steady improvement since the original MAP launch in June 2014 (this year will mark five years of nanopore sequencing!).  But it still suffers from systematic errors -- as noted in a recent Nature Biotechnology letter and some valuable comment from the original authors.  At the San Francisco meeting Clive Brown outlined a broad array of approaches to improve the accuracy -- software updates, new pores, use of nucleotide analogs, molecular barcodes and more.  To beat back Illumina's whispering, ONT needs to make regular progress on this front and demonstrate that progress with unimpeachable public demonstrations.

Of course the other approach to beating back Illumina is to play to ONT's competitive strengths.   One area where long reads are clearly having huge impact is in agriculture, where there are now many documented examples of copy number variants which drive important agronomic traits.  Mike Schatz's 100 tomato genomes in 100 days (with Sara Goodwin driving the PromethION effort) is an excellent example of this.  The San Francisco meeting ended with a talk on cannabis that demonstrated a complex CNV responsible for the balance of THC and CBD, which not only are pharmacologically different but affect the U.S. classification as marijuana or hemp.  So a CNV that has a major regulatory phenotype! Since many crop species have highly variable genomes -- and perhaps ill-defined boundaries with wild relatives which can be useful sources of new traits -- long reads at great scale are likely a winning story.

Another thing to watch for would be Oxford landing another really big project.  Their tie-up with Amgen counts as a first big project, but another big splash could really propel them.  Even getting just a pilot project for one of the emerging national genome projects would be a serious coup for ONT.

There will likely be other applications emerging where long reads, even with noise, can be as good or better than short reads.  For example, ONT (and others) have been reformatting Hi-C approaches to work with long reads.  Longer fragments mean higher mappability and packing more tags into one read can give more information -- every pairing of reads implies chromatin adjacency.  Single cell long read transcriptomics?  Another emerging area to watch.

Extreme speed to a useful result is another advantage of ONT over existing competitors; more applications need to be found that depend on this.  Low capital cost and modest running costs are important factors as well -- applications for these also need development.  To give an example from a recent Twitter conversation, increasing numbers of high school and college classes are performing DNA barcoding experiments, typically with the students amplifying from food samples and then sending out for Sanger sequencing.  If ONT wants to convert those classrooms over to MinIONs, they really need to make sure teachers have a plug-and-play solution from end-to-end -- particularly on the compute side.

Which also brings us to Flongle, the scaled-down, simpler to use flowcell for the MinION and GridION devices.  The few users who have gotten devices have been happy; can ONT quickly scale these up to full production without hiccups?  Education is just one of many markets where Flongle could spur great growth.

Getting into the clinic is a priority for any company selling biological analysis capabilities.  Each time I go to a nanopore meeting, it seems another good application surfaces -- pre-implantation cytogenetics that avoids freezing embryos, rapid diagnosis of infections to match antibiotics to bugs*, rapid diagnosis of cancer fusions to match tumors to drugs.  Some of these not only have the speed advantage, but also long reads are inherently better than short.  In particular, fusion gene detection is hard with short reads but easy with long reads, simply due to an easier time unambiguously mapping the long reads.

Really being in the clinic is hard.  Devices, software and reagent kits need to be appropriately vetted for regulatory purposes.  Diagnostic developers are going to want -- or more likely demand -- a commitment to a degree of platform stability in the form of frozen branches of instruments, code and kits.  Oxford, IMHO, really needs to make this roadmap very clear -- simply doing so in a public way would (I believe) help attract the attention of diagnostics tool developers who may not have considered the platform.


What about the rest?  Oxford always has many, many irons in the fire.  Some may have near term positive impact.  PromethION looks like it will launch commercially and should succeed; I haven't heard grumbling about bad flowcells for a while.  MinION MkIC -- combining the MinION instrument with the MinIT dedicated compute unit also seems ready to leave the nest -- along with the MinIT itself.  The few Twitter comments by MinIT users have been absolutely glowing.  The Cas9-driven target enrichment discussed in San Francisco seemed nearly supernatural in performance and embarrassingly simple.  Oxford has already rolled out their field sequencing kit, a version of the transposon prep requiring no cold chain.  Further improvements in library kits enabling lower inputs will always be welcome.

On the other hand, VolTRAX continues to be the sample prep of the future -- looks nice but seems to be a hard sell for real use, with too few kits available.  Converting it to a sequencing device is truly a wow! concept, but the incremental value over the existing platform may be small relative to the technical difficulty.  SmidgION sequencing-on-a-phone could be useful but I think MinION Mk1C plus Flongle really delivers for the space of ultra-portable sequencing capability. The Ubik no-lab library prep device is intriguing, but it would appear it is also highly inefficient and so will be useful in limited settings.

I'm sure I've forgotten as many things as I've mentioned -- that's a challenge keeping up with ONT (a truly smart commentator would make up a spreadsheet scorecard of every idea that tumbles out of Clive's mouth -- something to consider). Oxford Nanopore is never boring to watch; 2019 should prove no different.


* reminder -- I have a vested interest in this one as I am a paid adviser for a company in that space.