Tuesday, October 25, 2016

Messaging Counts

My recent piece on SeqLL's reboot of the Helicos sequencing technology attracted a number of helpful comments either on blogger or via email.  Several of which pointed out the high suitability of this box for counting applications such as expression profiling, ChIP-seq and copy number analysis. Julia Karow of GenomeWeb graciously provided me with a copy of her piece, which underscored the value of GW in having real reporters dig into a subject.  Much of what she dug up concerns SeqLL's future plans and for the near term mostly confirms suspicions that the technology will mostly look like the previous Helicos box, just scaled down.  But Julia also got out of SeqLL some of the numbers that are shockingly absent from the press release, such as the throughput (though I don't see the number of channels, a key aspect of this technology as covered below). My correspondents also pointed out a number of careless errors in the piece: I sliced $100K off the price of the PacBio Sequel (doesn't mean you can't try to get them to honor it!) and the In Sequence brand is long-gone from GenomeWeb's site.
Something emphasized by several correspondents is that counting applications are least hindered by the SeqLL/Helicos technology's shortcomings in terms of readlength and accuracy.  So long as things work well enough to get an unambiguous mapping, then short can work.  The simple sample prep is also a help in many counting applications, since these are likely to be performed on a large scale.  Most people (but not me!) are content to sequence a few genomes, but you'd ideally explore expression on many biological replicates across many conditions.  If copy number profiling is really going to hit the clinic in a big way -- I haven't kept up but my impression is that is still an open question -- then simple sample prep will pay there as well.  SeqLL, it was also stressed to me, can work on very small input amounts and damaged (e.g. FFPE) DNA.  This is why I'm surprised that SeqLL doesn't have a better description of their instrument's specs, particularly the number of channels on the flowcell, a key determinant of how many samples can be processed simultaneously.

Looking over SeqLL's site, I'm wondering if perhaps they should re-think, or at least tune, the levelo of emphasis.

In general, I think SeqLL would be wise to steer the focus away from sequencing and towards the emphasis on precision and accuracy in quantitation.   One change I would suggest would probably not go down well, but I will argue it.  The overall tagline for the technology is "True Single Molecule Sequencing".  For a long time, Helicos had the only single molecule sequencing; everyone else was sequencing clonal populations derived from single molecules.  But now with PacBio and Oxford Nanopore are on the scene, and Genia may be also, and they are also true single molecule methods with no PCR in (most) of the library preparation methods.  But all of these methods are superior to SeqLL in the type of data generated, which is why you don't want a customer asking "but doesn't PacBio/ONT/Genia have single molecule sequencing? How are you better?", because that starts straying into dangerous territory, particularly with (eukaryotic) RNA-Seq.  If someone starts asking about alternative splice isoforms, then the short reads off SeqLL are decidedly at a disadvantage to the long read methods. 

I'm also wondering if SeqLL should take more of a Southwest Airlines strategy.  Southwest's original target was not another airline, but rather (as related in many books) the bus lines such as Trailways and Greyhound.  The analogy here would be for SeqLL to gun after microarray companies and similar companies such as Nanostring.  

For example, compare SeqLL versus Nanostring.  Neither requires amplification.  Both are good at counting.  Nanostring requires synthesizing specific probes in advance; SeqLL does not. Nanostring is good for profiling 10s to about 200 targets in its most flexible mode (I think it is up to 700 using more expensive and longer-to-synthesize probes); SeqLL doesn't require declaring targets in advance.  Nanostring can achieve very low costs per sample, well below $100/sample (but dependent on number of probes and how many samples their synthesis cost is amortized over); it is unclear whether SeqLL can go that low.  Nanostring needs essentially no prep beyond lysis; SeqLL needs purified nucleic acids.  Both can handle DNA and RNA, but Nanostring would have the edge with prokaryotic samples, which lack poly-A tails and are dominated in RNA-Seq by ribosomal RNA unless it is removed, an expensive and sometimes troublesome step.  So SeqLL doesn't clearly beat Nanostring, but certainly has strong point.

Comparisons with microarrays will also often come out in SeqLL's favor. No setup cost,  no specialized consumables to stock.  No labeling.  Similar sample purification.  Microarrays don't get the glossy headlines anymore, but I'm frequently reminded that this technology hasn't gone away and often has cost-per-sample that keeps RNA-Seq and similar methods at bay.

Or how about qPCR and droplet digital PCR?  These are both heavily used technologies, in part due to low upfront costs but also due to widespread familiarity.  Simplicity also plays a role, but given the limitations of these methods they can serve as a gateway to wanting a broader look.  I'm more familiar with qPCR operationally, and it just doesn't scale well for lots of genes.

If we look at the press release for the new system (which curiously, isn't highlighted on the initial SeqLL webpage) in detail, the emphasis (as I covered last time) is all about sequencing and other sequencing systems.  Now, partly this is because they are apparently seeking only a small number of beta testers; this isn't a mass trial like Oxford Nanopore's MAP. But if SeqLL is looking to launch a system next summer, which is also apparently the case, then they need to start thinking about marketing strategy.  As I suggested in the prior piece, I think that is really going to require focusing on a small number of techniques -- say eukaryotic mRNA counting, ChIP-Seq and copy number analysis.  

Then the question is what market to target.  SeqLL is clearly aiming to try to separate current users from their Illumina and Ion Torrent boxes, but I'd suggest the alternative strategy of trying to convince people to upgrade from their microarrays and quantitative thermocyclers. I suppose the choice depends on which you think is an easier convincing to accomplish: move from a fading technology (microarrays), from low plex to high plex (qPCR/ddPCR) or whether there really is a large pool of users unsatisfied with the precision of their quantitative assays on clonal sequencing systems.  I doubt anyone has a crystal ball for this, but considering that the original Helicos failed to make sufficient traction with the "better than clonal" argument, I would argue that the other markets are likely to be easier to penetrate. As with all sequencing platforms, it will be fun to watch this play out whichever path SeqLL takes.

No comments: