Illumina has been branded a monopolist by two different governments, due to a worldwide lock on at least 80% of the sequencing market and perhaps in the high 90s in the U.S. and Europe. They've also had some disappointing quarterly reports, due to a flattening in their existing markets. Being forced to abandon the PacBio acquisition must sting. So where to next?
One challenge is that if growth in sequencing has flattened a bit, that wouldn't be the most opportune time to expand the capability of the NovaSeq. It's also not clear whether much can be done within the current instrument, other than extending the 2x250 mode beyond the SPrime flowcell. There is the risk that BGI/MGI will start being serious, but that company keeps channeling Sarte in terms of launching into the U.S.. We're told they will and that they've raised money to do so, but the trigger has yet to be pulled.
Illumina's key challenge is to grow without discouraging others to invest in ventures which ultimate ship capital to Illumina. Some prior business expansions have run head up against other competitors. Now, since for really big projects Illumina is effectively the only game-in-town one might think it doesn't matter, but if Illumina is too frequent in this the venture crowd may steer money away from the sector rather than proving a concept only to see their lunch eaten by Illumina.
Illumina could also expand their reagent and ancillary equipment offerings, though the trend has tended to be to streamline their reagent and kit lineup and just ditch the equipment; NeoPrep's failure perhaps casts a large shadow. Single-cell sequencing is exploding and may become a growth driver, but perhaps they are content with just reaping the increased sequencing demand that drives
In thinking what they might do next, I did come up with something that is both evolutionary and might drive further demand. It might also be technically near impossible, but that's where my lack of knowledge gives me sufficient confidence to throw this idea into the open (a rare case where Dunning-Kruger might be useful, or at least entertaining)
I wrote recently on the idea of having human genome sequencing races to drive speedier turnaround of clinical results when that can really matter. I'd love to see an Illumina Racing Team in jumpsuits hot-rodding a NovaSeq for the sheer spectacle, shaving down reagent delivery times and imaging scan times. But then I thought of another, more mundane reason to trim sequencer run times: the people who tend and feed sequencers like to keep normal schedules.
Below is a plot of NovaSeq run times from the specifications plotted versus the number of cycles. The three lines, from bottom to top, represent SP/S1, S2 and S4 flowcells. All of these are paired-end; Illumina doesn't sell any single end kits for NovaSeq.
In this plot, it would seem reasonable to say that the extrapolated time for 0 cycles represents the fixed components -- clustering, P5/P7 barcode reading and flipping over to prepare for the second read. The slope we would think is time per cycle.
From this we first see that SP and S1 flowcells lie on the same curve; the SP doesn't save time only reagent usage (and bucks!). The S2 flowcell has a somewhat curious change in slope. We also see that while the SP/S1 and S2 flowcells look like they'd have roughly the same constant component (7 hours) that isn't where the S4 line would track to.
But I'd particularly like to call attention to is that there are three kits with a very awkward run time of 25 hours: 300 cycles on SP and S0 along with 200 cycles on S2. That's a troublesome run time since it is one hour longer than a day.
Imagine you are running a sequencing facility that has a single NovaSeq and often has S2-scale runs. NovaSeq is a dual flowcell instrument, so on Monday you fire off a run at 1pm. At Tuesday you use the other side to fire at 1pm and then around 2 clean up the first side. Thus repeats the week.
But suppose management says: wait, why is my very pricey instrument only half utilized. Well, not exactly half, but basically so. And lets suppose that there is some reason to not want to go to an S4 -- perhaps the data is needed sooner than the additional 11 hours to run 2x100 on the S4. Or perhaps you have different library processes from different vendors with incompatible barcoding schemes (well, perhaps not completely incompatible but confusing to put together on same sample sheet). Or some other sound business reason to keep running S2s.
Well, if we start both on Monday at 1, then they both finish at 2 on Tuesday. Which means the next run probably won't start until sometime before 3 and finishes around 4 on Wednesday. Now someone needs to load them at end-of-day to get them running into Thursday, when they finish around quitting time -- and Friday will be worse.
Of course, one could launch Monday earlier. But having your runs keep marching up the clock isn't ideal. And having it come exactly in at 24 hours probably isn't great either, because there's no room for silly delays like being interrupted. So let's set a target at reducing the time for running 300 cycles, i.e. 2x150, on SP or S1 down to 23 hours, 2 hours shorter than the current spec.
There'd obviously be value in reducing the fixed time. But suppose that can't be touched; what about the cycle time? Using the logic above, I find that the slope is generally 3.6 minutes per cycle and that to shave two hours off we need to get that down to 3.2 minutes per cycle. That's only 24 seconds or 11 percent reduction.
How to achieve that? I haven't a clue because I don't understand the instrument at that level. Probably almost nobody does who doesn't work for Illumina or has tried to build a similar instrument. Faster imaging? Tweaking some fluid movement? Shortening incubation times?
But the payoff could be significant. Having a more convenient run time means more temptation to load both sides frequently, which means more sequencing kits used. Better utilization means easier justification of additional instruments, which must be fed.
Now, as a bit of transparency, my math doesn't quite make sense for other flowcells, so maybe there's a flawed assumption. Actually, even the high end of the SP is a bit longer -- it takes 3.9s per cycle by my math for the 2x250 SP flowcell. Curious. But worse is I get longer numbers per cycle for S2 than S4, which doesn't make sense. Now there may be some fudge in the Illumina numbers: they don't claim those are exact.
None of the other timings come out so awkwardly -- the longest run is 44 hours for an S4 300 cycle flowcell. But improvements in cycle times would look good across the board, and would of course anyone in a race-against-the-clock.
Just to revisit that fixed component, it is interesting that the shortest MiSeq paired end chemistry (2x25 using v2 chemistry) is only 5.5 hours, shorter than the 7 hour estimate for the constant component. Now, MiSeq does use bridge amplification instead of exclusion amplification, so that could explain some of this. But certainly a 7 hour fixed time operation would be a great place to optimize.
We'll see on Monday at JPM, or perhaps at AGBT in February (and that I'll be attending!) whether Illumina is thinking on similar lines.