Monday, May 19, 2025

Clive Brown At ONT: A Belated Retrospective

on Calling is imminent, and this is a notable one: the first held without Clive Brown in an official capacity at Oxford Nanopore. I started drafting a piece on Clive's tenure at ONT as soon as Nava Whiteford first broke the rumor (soon confirmed) he was leaving - that was back in November.  But first there was writer's block and then a pair of elderly relatives had health crises, only one of which resolved desirably, and then the piece got stuck in my procrastination queue.  But London Calling was an absolute deadline I set for myself and here we are.

One contributor to my discarding early drafts was trying to set the right balance.  Clive was by far the most visible leader at ONT and so to him goes much kudos but also much criticism.  Of course, in many cases it was someone else who deserves the credit or the debit, or at least the story is complicated.  But I was never, as the song goes, in the room where it happened, so I am sadly blind to such nuance.  It's also the case that Clive of early ONT is almost certainly not Clive of late ONT, and some of the foibles I dredge up might not be repeated.  But, they are part of the story.

It cannot be over-emphasized that under Clive's technical leadership Oxford Nanopore condensed an incredible (as in, not to be believed) concept into an incredible (as in, OMG!) working sequencing technology. ONT sequencers enable sequencing about any place on this planet (and even above this planet!) a human can go due to the compact MinION design as well as its lack of moving parts, have stunningly low capital cost and are simple enough that even middle school students can be trained to sequence.  The early "barely can align" data quality has advanced by leaps-and-bounds so that many reads have error rates of under 1%. Data yields early in MAP were low tens of megabases; MinION flowcells now deliver a few tens of gigabases and the PromethION flowcells significantly more.  I was hooked seeing a full length lambda phage of 48 kilobases in our first MinION Access Program (MAP) run; the world record ONT read length is almost 100-times that - and no other technology reliably generates high accuracy reads of even 48 kilobases.

Clive's team did that.  He built the technical organization to make magic real.  Now, it must be said that being a member of that organization came with special requirements.  Some staff stayed a long time - but some did not.  At the first Nanopore Community Meeting at the New York Genome Center, ONT had downplayed the idea that Clive would break any technical news - and all of us customers were madly scribbling notes and photographing slides as he blatantly violated that guidance.  When it was over, one looked around to realize that ONT employees had done the same and some very much had "deer in the headlines" faces.  Indeed, mutterances of "first I've seen of that" or "there goes all my timelines" were heard afterwards.

Eventually, ONT settled on a solution to allow Clive to be minimally restrained but the company minimally exposed to his wilder extemporaneous comments.  After Clive's talk (which led with a disclaimer that he might color outside the lines), Rosemary Dokos would be the voice of reason and carefully shepherd expectations into what the company was actually committing to

Under Clive's leadership, the MAP had some interesting aspects as well.  Clive has always projected a vision of the individual biological pioneer, in many cases an amateur with no training or background, casually using ONT gear to explore the world.  A parallel is often drawn, often dangerously bordering (or crossing into) fetishism, to the early days of personal computing.  I know that era: my first taste of electronics was learning the color codes on resistors so I could sort them upstream of my brother and father assembling a single-board DATAC-1000 computer.  But do-it-yourself can collide with amateur.  When MAP rolled out the first, a thinly documented, version of MinKNOW, I started looking for what seemed obvious to me must be in the package.  MinKNOW would output data in an HDF5-based POD5 format, and obviously there must be another tool to extract the reads as FASTA or FASTQ from the POD5.  Right?  Right? Surely????

No, there wasn't.  If you wanted to, y'know, analyze MAP data your had to dig into some very unfamiliar software guts.  In my case, Perl was a dead end because the HDF5 library choked on the POD5s (by 2014 I was realizing Perl library development and maintenance was pretty much in zombie mode; around that time the PDF module in CPAN had syntax errors!).  Since I wanted to learn Julia, I quickly tested if it had a working HDF5 library and wrote a simple parser. Several years later I found out it didn't work anymore for the same reason I had been reticent to open source it publicly - the implementation was tightly coupled to the POD5 implementation.  Presumably the same issue befell Mick Watson's R solution and Nick Loman's Python one, though by then ONT had made FASTQ extraction a standard part of their pipeline.

Similarly, ONT for a long time avoided writing much in the way of documentation or having much in the way of technical support.  You threw your questions to the Nanopore Community and you'd often get answers.  But as the Community aged, it became very hard to tell which information was still valid and which was badly obsolete.  Which version of a protocol did the search find?  Who knows?

What I said about open sourcing my extractor leads into another big tension area: openness and secrecy. I knew ONT was sensitive about internal workings and so asked if I could release my extractor, and got a not emphatic reply that suggested they wouldn't love that.  Mick and Nick must not have asked, as they did open source theirs.  For a long time, much of the software from ONT was only available from the Nanopore Community site, which was far less convenient to download from than GitHub or similar.  Over time that went away.  In general, ONT has been very open which has helped drive much innovation by the community, but now is probably enabling BGI and other Chinese competitors who are launching similar nanopore platforms.  Of course, ONT in turn benefited from PacBio's great openness and particularly the long read software community that grew up around PacBio.

From secrecy we can slide over to an entertaining topic: Clive and ONT's penchant for combativeness which often edged into destructive corporate paranoia.  In this he wasn't alone: CEO Gordon Sanghera and majordomo Spike Willcocks would also engage in this behavior; I chided Willcocks here on an egregious case that threatened to spike customer relationships. If you talk to the older ONT crowd, they have some very colorful stories to tell of then Illumina CEO Jay Flatley's behavior around them when Illumina had an investment in ONT (and yes, I'd love to hear Flatley's or any other Illumina's folks take on this - as off the record as they would like).  Illumina would later yank their investment in ONT and then try to sink ONT with a patent lawsuit. One can understand the bitterness after such a scorch-the-earth style of corporate battling.

But some of that activity, and the lawsuits with PacBio, brought out behavior in Clive that must have put the entire ONT legal team on high doses of ACE2 and proton pump inhibitors.  Clive would tweet out very sharp commentary on one of ONT's legal opponents, often with insults.  This would escalate a bit, until finally Clive's Twitter account would be "deleted'.  There would be a pause for weeks or months then Clive would start tweeting innocuous stuff again - which would eventually get spicer. Rinse and repeat.

In a similar vein, I'll always remember the first time I met Clive face-to-face.  It was at AGBT in 2013, a year after the big AGBT splash.  I thought that presentation was exciting, but apparently Clive and other ONT folks were run through the wringer of criticism as purveyors of vaporware, hucksters trying to just fleece gullible investors and so forth.  It had really gotten to him.  We met in one of the little outside alcoves that existed at the old Marriott facility in Marco Island and he showed me a MinION - probably one revision back from what came out for MAP.  And my big takeaway from that was "man is that guy strung tight!".  He was intense - both the great kind of someone who is passionate about their work and the less desirable intensity of someone who feels hounded.  The next time I saw him, when I helped him navigate to the ONT demo that fall in Kendall Square after bumping into him on the street, he was a bit more relaxed - building towards success can do that.  But still passionate - always passionate.

Clive's passions and ethos of the individual explorer sometimes took the company in directions that were of questionable commercial relevance.  Thrilling to watch or ponder perhaps, but not much in the way of a path to profitability.  For example, last year I was certainly excited at Clive saying ONT had succeeded in sequencing some of the smaller yeast chromosomes as a single fragment.  But I work at a strain factory that loves to work on yeast; few are in such a situation.  And yeast chromosomes are a tiny fraction of the length of even the shortest mammalian chromosome. 

Perhaps the most egregious case of this was the "Ubiqibopsy" demo.  Live, on stage Clive had his cheek swabbed, then some noisy bead-based sample prep, a rapid prep and by the end of the presentation ... a handful of reads.  Supposedly they proved Clive is human though I don't believe any reviewer #3 got a look at the data.  The device was visible later and clearly hacked together from ubiquitous lab parts - if I recall correctly there was part of a centrifuge tube, a micropipette tip and maybe one other recognizable bit.  But was this the path to a real product?  Given the challenges involved in getting to a data yield that might be interesting, unlikely - and little was heard of this ever again.

But of course nobody though the whole concept would ever work.  But then again, while ONT has shown technical success it still struggles with financial success. 

Similarly, Clive's love of the VolTRAX electrowetting technology was never returned by that tech.  In one talk he presented the idea of putting the nanopores actually on the VolTRAX chip, which certainly solves the transfer problem from VolTRAX to flowcell.  But the number of pores was to be tiny - so what application wanted a few hundred dollar prep to get maybe a few megabases of data?  That was a question never answered - and it too disappeared from future presentations.

At last year's LC, Clive gave hints that he might not be at ONT by the next year - something along the line of "if I'm still here".  There were other signs - the ElysION (formerly TurBOT) robot is the epitome of what Clive disliked, a large expensive box specialized on a single task and marketed to large faceless labs.  

I hope Clive is enjoying his retirement.  His Twitter feed shows very brief life very rarely, but mostly about non-sequencing topics.  If there isn't a third act for him (he was on the team that brought Solexa's sequencing technology to life), then we all now have a name to scan for in the annual King's Honors List. Under Clive's leadership Oxford Nanopore launched a truly revolutionary platform which has delivered huge gains to genomics; we should all be very grateful for that.

Wednesday, April 23, 2025

AGBT Flashback: Scale Biosciences’ QuantumScale

Back at AGBT two months ago, Scale Biosciences CEO Giovanna Prout was kind enough to spend thirty minutes of her hectic schedule with me discussing Scale's QuantumScale technology, which embarrassingly I've let those notes be sucked into a maelstrom of procrastination.  I was reminded of my delay by the announcement this week of Scale partnering with the Chan Zuckerberg Institute (CZI) to apply the technology to CZI's "100 Million Cell Challenge" and planned "Billion Cells" single cell expression profiling projects.  The QuantumScale technology is now fully commercially available.

QuantumScale has eye-popping specs.  It is possible to profile 4 million cells in a single experiment, with those cells divided among up to 9216 samples using ScalePlex technology, which I covered last June.

Scale has partnered with Ultima to enable Ultima indexing kits enabling native Ultima libraries, rather than requiring a conversion step.

A quick review: Scale's approach to single cell profiling avoids any specialized encapsulation equipment ala 10X Genomics by fixing cells so that the cell remnant serves as the addressable unit for indexing reads.  After labeling, pooling a sample and then splitting it again enables combinatorial barcoding.

Library preparation starts with the option of using ScalePlex to index multiple samples.  Fixed samples then are processed with reverse transcriptase and indexed on the first barcoding plate.  The entire set of indexed samples is then pooled and added to the QuantumScale barcoding plate.  QuantumScale, as I covered in September, doesn't just add a well=specific barcode to each new split of the pool, but instead each well has the capability to index at much higher, well, Scale.  In the Fall Prout was keeping quiet about the exact means QuantumScale used to achieve this, but at AGBT she revealed it relies on indexing beads - 800K per well.  The entire workflow can be completed in a day and a half.  And she is confident the technology can go to even higher levels of indexing; they just need sequencing costs to drop further to enable such plans.  Though right now the cost per cell is quite good - library prep costs of 0.8 cents/cell for 4M cells or 1.0 cents per cell for 2M cells or just $5K all-in to profile 84K cells . She also noted the entire protocol - including the microbeads - is very friendly for liquid handling robots. 

Prout also mentioned that Scale is developing a probe-based workflow for Formalin Fixed Paraffin Embedded (FFPE) samples, the dominant sample type of large tissue archives but also the nightmare of molecular profilers as the FFPE preparation process and the deparaffinization tend to both damage and break any nucleic acids.  It actually could be worse: very early FFPE protocols used unbuffered formalin and tended to completely destroy nucleic acids.   

I can still remember the first stirrings of single cell sequencing about a decade and a half ago, when it truly meant "single cell sequencing" - data was generated for individual cells in wells.  We're so spoiled in genomics by routine exponential climbing of the price/performance curve, but its still astounding to contemplate profiling 4M cells and potentially dividing that largesse over a matrix of over 9K different conditions and replicates.  But that is the corollary of those fantastic growth curves: what is barely imaginable today will be fodder for proof-of-concept in a couple of years and then routine a decade later.  And with the growing power of machine learning models, the demand for such ginormous datasets seems to be on a nearly vertical upward curve.  QuantumScale is emblematic of such growth of genomics capabilities, now we can sit back and see what biological insights tumble forth from the experiments it enables. 



Thursday, April 17, 2025

Stellaromics Dives into the Thick of Spatial Genomics

2025 has sometimes seemed like an unending dark winter of biotech, but just before AGBT spatial genomics company Stellaromics announced they had secured $80 million in Series B funding to advance their Pyxa platform.  Around that time, CEO Todd Dickinson and CTO Ye Fu chatted with me by phone, which any diligent and responsible scribe would have written up immediately.  Alas, that is not me.  But even two months of delay in writing cannot change the fact that Stellaromics is an interesting new entrant to the spatial genomics field.

Monday, April 14, 2025

Will the Result Be GeneXpertION?

One of the more intriguing and slightly mysterious genomics news stories of last week is Oxford Nanopore announcing a partnership with Danaher's Cepheid business to use Cepheid's GeneXpert system for sample preparation upstream of nanopore sequencing.  The press release was filled with grand visions but few details, which isn't exactly unusual for ONT.   I've often heard of Cepheid but never really dove into the technology, so I've been doing a crash dive on the company's website (which was also a raiding expedition for the images used below) as well as a detailed post Nava made in 2022.  Here's what I have learned and also some speculations about how it might fit with ONT - and if I've misunderstood anything I hope to receive constructive guidance on where I've gone wrong.  Please note I've made a COI disclosure at the bottom of this piece.

Tuesday, March 18, 2025

Mission Impossible: Methylomics

Good morning Mr. Hunt.

Today's briefing will have a bit more background than usual - you haven't tangled with biotechnology since that odd little company in Australia with the fancy office tower and actual labs in caves.

As you may know, DNA has four letters or bases which form pairs; A with T and C with G.  It is also possible for the C to be modified by methylation to form 5-methyl-C or even 5-hydroxymethyl-C.  These in humans are always at C followed by G, called CpG, sequences.  There is great interest in reading the methylation of DNA from blood, as this "cell-free DNA' may be a oracle into current and future health conditions.

The best technologies for reading this are the single molecule sequencers from Oxford Nanopore and Pacific Biosciences, as they can read these marks directly with no additional preprocessing of the DNA beyond what is required  by the sequencer to just read the bases, the construction of sequence libraries.  But these suffer from relatively high input requirements, and any amplification of DNA by PCR or similar techniques erases the methylation.

It is possible to read methylation on the popular short read sequencers of Illumina and other companies but only with a trick.  The most popular method is to treat the DNA chemically with bisulfite, a rather nasty reagent; even with your disdain of danger, you really should read and adhere to the MSDS on this stuff.  It converts all unmethylated Cs to Us and so it pairs and sequences like a T; modified Cs are untouched.  Please do not bring up this technique with the bioinformatician joining your team; they are known to rage about bisulfite being "a weapon of mass sequence destruction".  Similar methods using enzymes produce the same result.  Conversion means this DNA can be amplified.  But it also means it is useless for calling genetic variants; a separate unmodified library must be used for that.

Watchmaker Genomics has a clever chemistry that performs a much more limited transformation - only methylated Cs are converted to Us.  An even more clever biochemistry is offered by Biomodal, which copies one strand of a DNA fragment into a second, linked strand and then treats with enzymes.  After sequencing both linked sides, the pattern of matching and mismatching between the two can call variants, 5-methyl-C and 5-hydroxymethyl-C.  

Roche has recently unveiled a new single molecule sequencing technology called SBX, but it requires first copying the DNA of interest into bizarre highly modified "expandomer' form.  So it shouldn't be able to read methylation.  But at AGBT, SBX boffin Mark Kokoris offered, in response to a question on the topic, that there was clearly room for a fifth signal level in their traces and hinted Roche was working on using that fifth level.   But how, if SBX requires copying the input DNA first?

We first thought of sending you in to extract the secret from Roche, but even we can't just go performing espionage on a legitimate company with no apparent plans for world domination; monopolization of the sequencing market by Illumina has never triggered us to action so that isn't a justification.

Instead, your mission, should you choose to accept it, is to realize the impossible by creating a system for creating SBX molecules using a third basepair, and such artificial basepairing schemes have been realized in the lab, so that the set of all 6 bases can be used to resolve A, C, G, T and 5-methyl-C; we will save 5-hydroxymethyl-C for a future mission.  Ideally this would be achieved by specifically converting 5-methyl-C into a base of the third basepair type, but if you devise schemes converting C or T to the extra basepair and then 5-methyl-C into whichever base you freed up, that's acceptable as well.

Our technical directorate has outlined two general strategies which could be independent or used separately.  

In one, a purely chemical approach conversion would change one of the bases to the new basepairing arrangement.  No chemical reaction has ever been proposed in the literature to do such a transformation of one base into an unnatural basepair.

In the other plan of attack, protein engineering would be used to generate one or more enzymes to perform the transformation.  No enzyme is known that would provide an obvious starting point for such a protein evolution campaign.  Perhaps one of the new machine learning models can take a crack, but this is far beyond anything demonstrated with AI-based protein engineering.  

Either way, in the event of success you may be required to disguise yourself when presenting the results at AGBT.

A note about commercial considerations, and we don't mean the usual of making sure you hold your beverages with the label facing outwards.  Whatever process you devise must not have an inordinate number of steps, should not require a fume hood and the cost-of-goods should not be exorbitant.  Nor should your spending during the project; if you should exceed your budget authority the secretary will disavow all knowledge of your 

This message will self-destruct in 5 seconds - or would have if we hadn't spent the micro-explosives budget on DNA sequencing reagents.
 

Saturday, February 22, 2025

Is Midi Read Sequencing A Thing?

Throughout the two pieces on Roche SBX sequencing I sprinkled the term “midi read sequencing”.  Here I’m going to explore in more detail this concept.


Thursday, February 20, 2025

Roche Ripple Predictions

In the prior piece, I covered the technical details unveiled by Roche for their SBX technology, but generally tried to avoid predicting its effects on the marketplace.  Here I put on the pundit’s hat.  The TL;DR is this is a major new sequencing platform and if you’re at one the competitors you have about a year before it fully hits the market - though in reality the action has already started as Roche starts grabbing hearts-and-minds.  What can we anticipate about the effect on each of the current players? As noted in the prior piece, some key aspects - in particular purchase price and run cost - aren’t being disclosed by Roche and complicate prognostication.  

Roche Xpounds on New Sequencing Technology

Bar bets can be a powerful force in human society.  One of the best known books on the planet, The Guinness Book of World Records, originated from the need to equitably settle wagers.  Many entries in that tome are questions of immense scale - the largest this or heaviest that.  Shortly before this posted, Roche unveiled a sequencing technology that per its inventors may be the result of such a bar bet: how large a dangling bit can you stick on a nucleotide and still have it incorporated by a polymerase.