Monday, November 05, 2018

Illumina Buys PacBio: More Thoughts

Illumina surprised pretty much everyone in the genomics community by announcing the purchase of Pacific Biosciences.  I had spent Thursday deep in the weeds of a combined PacBio-ONT-Illumina dataset, so was caught completely by surprise on my commute home by an email asking for my comment.  If you do want to hear hot takes on it from myself and AllSeq's Shawn Baker, Theral Timpson over at Mendelspod interviewed us that night.  There has of course been much discussion of the deal and tributes.   I've had the weekend to ponder things, and here are some somewhat better thought out and detailed comments -- though I don't believe I've retreated from any of the themes in the podcast.  I've grouped the thoughts into a few themes.

Pacific Biosciences' Brought Forth Amazing Science

It can not be said enough what a pioneer Pacific Biosciences was or how much amazing science they brought forward.  The instruments rely on zeptoliter wells which are illuminated by light whose wavelength is bigger than any dimension of the well.  The whole approach involves continuously monitoring individual DNA polymerase molecules as they synthesize DNA.  As an undergraduate I once had a discussion with a favorite biochemistry professor circa 1990 and I complained that sequencing was too slow -- he countered he was amazed how fast it was, since he had seen the very early days.  Imagine if I had claimed that in under 30 years we could watch polymerases work?  I think he would have sent me to the infirmary!

PacBio had to fight intense skepticism on the utility of their data, given the high inherent raw error rate.  Their data didn't look like Sanger data and it didn't look like the Illumina data that was dominating, so too many people wrote it off.  But PacBio hired great scientists and gathered a stellar array of academic collaborators, and these folks published paper after paper to provide increasing value from the data.  Of course, some of that was that the data kept getting longer. Later would come complete genomes of large eukaryotes such as human, then haplotype-resolved genomes.  And remarkably, PacBio made all their methods open source, endearing themselves to the community -- though also enabling the competition.

 My personal experience is an attempt to use PacBio in Spring 2012 yielded 3kb insert libraries that didn't solve our problems when put through hybrid correction -- because the repeats giving me fits were much larger.  But just a year later, we tried again with the same vendor and got 6kb inserts -- and the reattempt was fueled by the excitement around the direct assembly (no Illumina error correction up front) that PacBio and collaborators had shown off at AGBT.  And it worked!  Most of my nasty repeats yielded, and we started throwing genome after genome through the pipeline.

Some personal reflections on the science by those who did it have been popping up -- Jason Chin, Andrew Carroll.

Be Glad You Didn't Buy and Hold From the IPO

While PacBio did great science, they struggled as a company.  Matthew Herper pointed out the grim reality: while Illumina is paying $1.35 billion in cash for PacBio and that represents a 71% premium over the average price for the previous thirty days, it valued the company at half the IPO price.

Particularly if it isn't part of your compensation, it is easy to say "who cares about Wall Street?  Keep giving me the science!".  Unfortunately, real companies do have to worry about Wall Street.  PacBio was continuing to lose money, with a rough runway of a year and change when they were bought.  Of course those losses might have narrowed or they could have found other ways to raise money -- strategic partnerships (like Oxford Nanopore just executed with Amgen) or new dips into Wall Street.  But being in that situation meant that management couldn't go wild investing in new applications

In particular, getting into the diagnostics realm has been the ticket to prosperity for many companies. Because of the highly regulated nature and the challenge of getting reimbursement, it can be a hard market space to get into.  But conversely, once you are there medicine has many conservative trends that keep technologies in place -- and all those barriers are now barriers-to-entry for your competitors.

PacBio had teamed up with Roche Diagnostics on clinical applications, and then had the rug pulled out from under them.  That's still a bit of an enigma given that Roche hasn't launched the Genia nanopore sequencing technology.  But it certainly was a blow to getting PacBio into the clinic. 

Illumina's has enormous financial resources at their disposal.  Their market cap is over $44 billion -- well over 50 times bigger than PacBio's before the announcement.  Illumina can use debt (they have a bunch out) or cash extracted from their nearly $200 million a quarter after-tax profit.  That leads us to...

[sid note: all of my family's investments are in broad mutual funds, save two blue chips that I've held since before that conversation with my biochemistry professor and my holdings in my employer and advisee company that are part of my compensation.  I don't short stocks, play with options or engage in any other financial practices which take advantage of short-term trends in the stock market or which could in any sense benefit from me commenting on a company] 

What Will Illumina Do With the Technology?

First, Illumina will need to integrate PacBio with their organization.  Unfortunately, that is always accompanied by a bit of brain drain.  Some people will find themselves holding positions that already exist at Illumina and some folks will simply want to explore new opportunities.  Illumina has integrated a number of acquisitions over the years and is probably quite good at it, but there is always some degree of confusion and unproductive disruption.

A key logic of the deal is that Illumina brings more resources to power the SMRT technology into more use.  For the clinic, Illumina has both the money to get things moving and extensive existing connections to key leaders -- not that PacBio had none, but Illumina will certainly have more.  And Illumina must have a bigger salesforce which will pester pursue more scientists in hopes of getting sales.  And anyone who held back because they were worried that PacBio might be a company without a long-term future can now cross that objection off.

Illumina will likely increase the investment in boosting the instrument itself.  PacBio was starting to get to be only a small multiple in price for a human genome class project, and further improvements in the number of ZMWs per flowcell could have (and may well still) driven that even lower.  

Illumina also has a significant reagents business and so might bring new kits to enable new applications.  Adapting Nextera would be one obvious one.  That might enable faster/cheaper library production but probably with a significant drop in data yield unless some clever trick can put hairpins on the molecules.  

Buying PacBio is a bit of a change from many of Illumina's more recent purchases, which were headed directly into the clinical (Grail) or personal genomics (Helix) spaces. 

What C-Suite Residents Are Kicking Themselves This Week?

It's interesting to ponder who might feel stung by the deal in terms of missing out.  Was PacBio quietly exploring being acquired or did this come entirely from Illumina?  That might be in the merger document, though I haven't tried to get one.  That would be one argument why I should have a small holding in all the genomics players, as I'd get all their big mailings!  

My Dad's old employer GE is a general mess right now; GE Healthcare is supposed to be spun out as a standalone company and that probably doesn't favor making a bold acquisition move.  Thermo has a whole collection of sequencing technologies and would be an obvious home.  QIAGEN could have been another linkup, though they're probably mostly focused on the less than stellar market penetration of GeneReader, at least in the U.S.  If Roche Diagnostics had called, would PacBio have even returned the call? Or you could imagine someone in the medical space that isn't in sequencing wanting to use PacBio as an entry point -- maybe Philips or a similar company.

Competitive Landscape for Long Reads

The competitive landscape for long reads really has two pieces: there's true long reads and there are linked reads aka read clouds.  

For true long reads, there's only one competitor right now: Oxford Nanopore.  PacBio has a clear technological edge in overall data quality.  Raw PacBio has mostly random error and in circular consensus (CCS) mode that error gets very small (well, so they say -- I haven't played with a long insert CCS dataset yet).  Indeed, PacBio was increasingly touting the long insert CCS capability, as getting really long insert libraries is so tough and dependent on so many aspects not under PacBio's control.  ONT, particularly on the PromethION platform (if you can get flowcells; a user was lamenting the difficulty in getting flowcells to me recently)  on the other hand has a huge leg up in speed, overall throughput, cost per base, capital cost and getting truly astounding read lengths.  And of course nobody hauls a Sequel into the middle of a jungle or tries to run one in an airport lounge.  

Illumina's financial muscle might enable the SMRT technology to at least not fall badly behind by upping the density.  And if a desktop mini-Sequel can be built -- I have no clue if that's utterly absurd -- then Illumina could fund it.  But there's only so cheap you can go with a powerful microscope, so Pacbio never going to have the low upfront cost of a MinION, GridION or PromethION.  

Illumina will also clearly have deep pockets to throw at patent litigation, continuing whatever is still ongoing with ONT (I've lost track) or initiating new suits.

In the linked reads space, there is the established player 10X and then a number of companies trying to launch equipment-free versions.   For example, MGI's announcement of their NovaSeq-class instrument (which I need to write up!) included the launch of their single-tube Long Fragment Read technology.  There were several other publications this spring/summer on similar technologies -- including one from Illumina.  Those options all will be cheaper than getting true long reads on PacBio, but they'll never be as useful as long CCS reads.   Long-term I suspect all of these will become niche applications for very small labs and certain uses, with 10X becoming mostly about single cell RNA seq and single cell immunoprofiling.

If you look on Twitter, you'll find a full spectrum of opinion on the effect of the deal of sequencing prices.  Some see Illumina driving the long read prices lower and that would clearly be a good thing.  Others fear that removing even a small competitor will reduce the pressure on Illumina to control reagent prices.  I tend to be optimistic, but can't claim a lack of disappointment in consolidation in the industry.  

Don't Write an Epitaph for Short Reads

And for anyone thinking this is the death knell of short reads, dump ice water on your head repeatedly until that idea goes away.  Many forces will keep short reads in business for a long time.  They're still cheaper.  The workflows are far better established -- simple institutional inertia will slow adoption.  And there's still many applications, particularly where the sequencer is being used as a generic measurement device and the payloads are tags, where short is not a problem and/or sheer numbers are critical.  And if you don't buy that logic, note that Illumina still has a thriving microarray business, laughing at the many (including moi) who though microarrays would long since have gone the way of Maxam-Gilbert sequencing.  I wish they didn't -- not so I could be write, but so I could stop grimacing at all the popular news items that have linked the merger to companies such as 23 and Me that use Illumina arrays but the news items identify zas using sequencing.


Dale Yuzuki said...

Thank you for the thoughtful post Keith - save me the effort of putting my own thoughts down!

Liang Zong said...

Thanks of these inspiring thoughts. If combining nextera and SMRT bell preparation together comes true, that would be magic.

Leeloo said...

What will be the best case scenario in terms of technical advancement in the sequencing technology when we empower Illumina with Pacbio's IP? Long reads that are cheap? More accurate reads because of SMRT bell incoporation into more things?

Bertrand said...

Hi Keith,
Thanks a lot for the super article (it's good to see you back, I was getting worried...). How about a quick piece on the "other" systems, namely Thermo Fisher/Proton, Ion S5 and Qiagen/Gene Reader? I tried hard to find some info on how they're doing (sales, number of systems installed...) with very little success. I expect they're probably marginal, but since this is a different (clinical) market there may be some surprises. Any ideas, or links to information?
Best, Bertrand (Marseille), unfortunate buyer of a Solid system a long time ago...

Anonymous said...

tagmenting smartbells is possible. our group already showed its possible. just getting long reads is difficult