PacBio Outlook 2018

Well, I didn't exactly get my Pacific Biosciences preview out before their J.P. Morgan presentation.  Luckily, the slides for that primarily projected financials and touted their successes -- and didn't drop any major platform announcements in -- so I didn't miss out.  PacBio's position is important and worth reviewing, even if it doesn't change much.
Pacific Biosciences was the first company to launch a successful long read single molecule sequencing platform.  Few in the community saw utility in high noise, ultra-long reads, so brilliant folks at both PacBio and their academic collaborators had to blaze those trails and develop revolutionary software tools.  PacBio also made the bold and laudable choice to make those tools widely available and open source.  For those working with their system, they've also wrapped everything up in a powerful bur reasonably friendly application -- and one that doesn't sacrifice command-line usability for GUI simplicity -- you get both.  

On the hardware side, the RS IIs remain in operation but the future is all Sequel.  So a simple lineup, in contrast to Illumina's crowded bench or Oxford Nanopore's array of sequencers and ancillary instruments.  So tracking PacBio's stable of instruments is simple, because you really need follow only one.

PacBio has developed several important markets for their technology.  If you want to use a single platform to sequence microbial genomes, then PacBio is still the gold standard.  They've shown that de novo assembly of huge, complex genomes is not only possible, but with long reads much of the haplotype and structural variation information can be recovered as well.  PacBio has demonstrated utility of barcoded circular consensus reads for replacing Sanger for sequence-verifying synthetic constructs.  Iso-Seq is gaining increasing acceptance as an approach to define the full length transcriptome of samples.  Direct methylation detection proves to be a way to parse metagenomes.  There's probably more; PacBio has been scrappy about finding niches ill-served by Illumina and Sanger that fit their strengths.

The challenge has been much the same for the last few years.  Versus Illumina, PacBio costs much more per raw basepair, so the company must convince users that there is significant value-add to Sequel basepairs over SBS ones.  They've demonstrated this many times for many applications, but still many haven't bitten on that argument.  At the other end, others are nipping at their heels on those carefully-developed applications.  

10X Genomics is mostly competition for complex genome assembly and haplotyping, though metagenomics could always be a direction for linked read technologies.  For the few comparisons I've found, 10X delivers much worse contig N50s but not radically worse scaffold N50s, but the data generation is much lower cost.  So investigators can choose a less expensive 10X option to deliver some fraction of their goals or the much more expensive PacBio to get more.  Sometimes the money might not be there, other times it may be more valuable to get a larger number of lower quality (but still quite good) assemblies than a few much better ones.  It all depends on your scientific goals.

But most of that heel-nipping is from Oxford Nanopore.  In the context of PacBio, those who get past the steep MinION learning curve -- and are willing to put up with ONT's lack of a unified, user-friendly toolchain -- find that the MinION can deliver similar data yields as PacBio.  The read lengths can be longer -- much, much longer -- but the data quality is decidedly poorer at the consensus level.  Despite the many improvements in ONT's base calling -- documented spectacularly well in Ryan Wick's preprint README file -- Oxford's consensus sequences hit a wall around 0.2% error rate but PacBio can get much, much better with sufficient depth. 

So MinION can't compete directly with Sequel for quality.  But if raw base quality might not be an absolute need, then that isn't going to be a strong deciding point.  Oxford's cDNA and now direct RNA kits are showing spectacular results -- and the protocols are shorter and less involved than Iso-Seq.  If you have a quality reference genome, then do you really need base-perfect full length cDNAs?  Selling investigators on that point is probably not easy.  Or there's the hybrid sequencing end-run.  And if you're going the hybrid route, then the required depth of long read coverage goes down -- and the difference between PacBio and ONT for the final results becomes much murkier.

MinION wasn't going to appeal to PacBio's focus market of core facilities, but now there's GridION which packs 5 MinION-like units in a single box along with key computational support.  GridION doesn't solve the downstream software issue, but it has been well-received and seems to have launched nearly flawlessly.  Oxford's inroads into markets may have been slow but steady.  For example, there isn't yet a commercial GridION provider in North America  -- and shipping samples overseas adds additional scheduling and documentation headaches that many (or at least I) would prefer to avoid.  

More ominously for PacBio, Oxford is again making noise that PromethION is moving forwards.  I remain a mix of cautious and skeptical and unconvinced PromethION wasn't a strategic error -- but perhaps it really will show up this year.  PromethION deserves its own piece, but a quick contrast can be drawn between users mentioning they still don't have PEAP instruments and Clive Brown tweeting out a photo of lots of PromethION chassis supposedly ready to ship.  Tweets of PromethION runs are rare, but are coming from more users.  Clive is now touting over 90gigabases per cell and mentioning the 48 cells per box numbers, but whether such full instruments are in users hands isn't clear and MinION users still tend to get about 20% of the yield Clive gets on his own numbers.  So PromethION is a wildcard, perhaps to drag out longer but always potentially dropping on the scene as a huge competitor.

So what's PacBio to do?  One is to keep the nose to the grindstone and continue to execute well.  Sequel went through severe growing pains -- I've seen stock analysts' report that really lays that out in detail -- but seems to have settled.  But the bigger need is to push performance higher.  Continual protocol tweaks provide a gradual lift, but the big jumps require major platform upgrades.  PacBio had promised a 2-fold jump at the end of 2017, but the one provider I've spoken with is still projecting the older 5 gigabase yields.  Each doubling makes arguing for premium quality easier, when the premium isn't as larger.  And each jump keeps Sequel on par or ahead of MinION/GridION for yield, at least until ONT decides to push that technology harder.

The other route to coping with competition in the research space is to drive quickly into clinical applications.  Clinical can be harder to get into, but also harder to be evicted from -- look at all the Sanger assays still being run.  Roche's ditching of PacBio about a year ago in this space remains a serious setback (*). PacBio hasn't made any super-noisy splashes in the clinical space, but has touted research results with potential clinical relevance.

So 2018 should be a year of PacBio continuing to grow while looking over their shoulder.  Perhaps nothing flashy or provocative, but plenty of  exciting science and perhaps at least one doubling of throughput during the year.  Perhaps some specific forays into the clinical space.  But mostly solid execution focused on the business of being a sequencing company.

[* - note that my day job now involves a major collaboration with another arm of Roche, in case you think that affects my objectivity]