Thursday, September 26, 2019

Where U.K.'s CMA Missed the Plot

I've written far more than I expected to on the PacBio merger, with it dominating my output here this summer save that recent dip in the CCS boomerangs.  We still haven't heard from the U.S. FTC, which is the big hammer that could drop and the stock is still selling at a substantial discount to the price offered by Illumina.  While I've hinted at it, I've never quite detailed my major objection to the form of the UK CMA's analysis of the proposed acquisition.  I've missed the opportunity to lay that out during the period of public comment for the next round of the CMA.  Still, worth it to write it down.  Which was true about a month ago when I wrote this, then failed to actually push it to a post before I headed out for a rugged vacation in the American West.  And then forgot about it after returning.  And more news today to comment on -- so better finally get this out there.
The CMA report contains many strange ideas and odd leaps of logic which I could try to enumerate in painful detail.  But instead, I'll focus on a theme that ties many of their missteps (from my perspective) together.  The CMA focused on the sale of sequencing systems -- as in instruments -- in their analysis, which IMHO meant going off the rails right out of the station.  What they should have focused on are consumables, as that's where the real market power of Illumina lies -- and where a lot of possible remedies for antitrust issues might lie.

Other than mongo sequencing facilities, most institutions buy instruments rarely.  They're huge capital buys.  They are also probably sold near cost, as they are huge, complex instruments which are made in very small production runs.  But this is what the CMA focused on.  

The converse would have been to focus on consumables and margins -- the flowcells and the chemistry required to run them.  These are unusual specialty chemicals and devices, but can be manufactured at a level where economies of scale show up.  Illumina likes to boast to stock market analysts, if not to customers, that their margins are robust.  That means that Illumina could, if forced to, sacrifice margins and profits to maintain market share.  It's also important to remember that if you are doing much sequencing, your consumable spend can very quickly exceed the outlay for bringing the instrument in house.  NovaSeq has two flowcells; start cranking S4s on both sides twice a week and you very quickly exceed the instrument sticker price.

It must be noted that since I have only seen redacted public versions of the CMA's documents, I can't be certain they didn't explore consumable margins.  But the pattern of the report writing makes this highly unlikely.  Typically there is a paragraph explaining a line of thinking with minimal redactions, followed by a highly (or often completely) redacted set of supporting bullet points or tables.  Since none of the report has such as paragraph on consumable margins, we can reasonably infer it doesn't exist. 

This line of thinking fouls the analysis in many ways.  For example, the CMA explored the idea that outsourcing sequencing is an alternative to buying a sequencer and that might affect the market.  It probably does in a tiny way, which was their conclusion. Once in a while someone decides to buy an instrument instead of outsourcing or vice versa.  But consider it from the reagent perspective: the question of outsourcing neatly becomes irrelevant.  Flowcells are flowcells, no matter where they are run.

The focus on instruments led to some bizarre conclusions.  Actually, I don't believe you can really blame the focus for that, because the conclusions are just so wacky.  The overall thesis is that the Sequel II is a competitor for the NovaSeq and this is an area for antitrust concern.  The supporting argument is that human (and other large) genome sequencing will switch over from Illumina to PacBio if the prices are similar and that the prices are similar.  Which is a take from bizzaro world.

A NovaSeq and a Sequel II occupy similar floor space.  I haven't gotten a solid price on a Sequel II, but it is perhaps 5/8 or so the cost of a NovaSeq.  But, in terms of throughput it is clearly a loser.  The recent PacBio human genome paper/preprint (which is very much worth reading) used 24 hour movie times and 12 hour pre-incubation of the polymerase complexes (I'm unclear if the preincubation ties up the instrument; probably not).  So roughly in the time one Sequel II generates a single human genome you'll get around 30 off the NovaSeq (with an S4 flowcell generating 3000Gb every 44 hours and two flowcells per instrument).  For the cost, I've seen estimates that it would take 2 Sequel II flowcells and those may be around $1K apiece (corrections welcome; haven't seen great numbers); I believe Illumina genomes are going from around $700 these days (again, corrections welcome).  Of course the PacBio data is superior, but CMA didn't discuss this nor is it yet obvious that many labs have the mindset or bioinformatics pipelines to take advantage of that edge.

That doesn't include the expenses for sample and library prep -- getting even 20kb CCS libraries means some care in sample preparation and certainly the library prep is far more involved than an Illumina Nextera prep.  And remember that the Illumina machine is good for a wide variety of counting applications that currently just aren't in scope for long read instruments.  There is motion towards multi-sequence-tag packing schemes ala the old SAGE RNA-Seq schemes, but realistically those are at least two years away from being reduced to general practice. In contrast, not only is Sequel II limited to a relatively small number of reads per run, but it also doesn't have the demand matching capabilities of NovaSeq.  With NovaSeq, there are different flowcells for different size runs; for Sequel II just one size (as I understand it, Sequel I flowcells can't be used)

Now in the context of the merger, we can sketch this out in the two scenarios: Illumina acquires or no acquisition.

In the Illumina acquisition scenario, Illumina has incentive to maintain the price difference.  The PacBio price probably won't drop much over time, because technically it will be hard.  Small improvements will come from evolving sample and library prep and perhaps informatics improvements and possibly even better loading efficiencies.  Really large jumps will require packing more ZMWs per flowcell, which means hardware redesigns and who knows what new issues cropping up.  Illumina could also cross-subsidize or bundle the platforms in different ways -- buy 4 NovaSeqs and get a free Sequel II!  Or if you maintain a steady buy of NovaSeq flowcells you get a sizable discount on Sequel II flowcells.

The no acquisition scenario is not rosy for PacBio.  Should Illumina every actually worry that customers are moving their jobs to PacBio, they can shave some margin.  They may start doing this anyways; BGI has been filing my browser windows with ads for $500 human genomes.  That may be a loss leader, but it still is the real competition for NovaSeq.  In any case, with their lower production runs PacBio won't be in any position to shave margin -- unless swallowed up by someone with deep pockets they must worry about keeping the lights on.  [note: big news on the merger today; I'll save that for a separate post]

Ultimately I expect large genomes to migrate over to long read platforms such as Sequel II and PromethION (if the latter really can fix the systematic error issues).  But I also expect that counting applications will swell so large that Illumina won't really miss this.  But the CMA's charge doesn't allow them to consider "they're losing X but gaining much more in Y" as an argument.

1 comment:

Anonymous said...

Fine to disagree, but you’re overlooking their definition of competition, overlooking that a 95% market owner is adding 2-4% from a competitor and/or potential competitor, overlooking that most of the [anti] competition arguments are drawn from internal documents from the parties themselves, the rest from surveys with actual users.