Tuesday, January 12, 2016

Illumina's Unveils Firefly

Illumina third big announcement around JPM is to unveil Project Firefly, a semiconductor sequencer which will use existing SBS library preparation and a derivative of SBS chemistry.   Slotted with a price point ($30K), physical size (small pizza box ish?) and data yield (4M reads, 1Gbp data)  below the just announced MiniSeq , Firefly would be two small boxes which could stack: one for library preparation and one to run single channel sequencing.  The flowcell would use ordered arrays, layered atop the semiconductor sensors.  Launch is proposed for the second half of 2017.

I think I've nearly covered the gamut of available details in my opening paragraph (though further scrutiny of the GenomeWeb article or Bio-IT article or other coverage could well find a few more nuggets) . Indeed, when I started googling after the announcement, one of the top hits was to a laughably mangled OCR that Google reported as "Firefly Meatless Illumina' (it was an old news item on the heatless illumination from fireflies

 I haven't grown entirely cynical on launch dates, but "second half 2017" means that any devices installed by December 2017 meet the deadline.  So let's call that two years from now: anyone else trying to establish a foothold in the low capital sequencing market (e.g. Genapsys) has two years before King Kong actually starts tromping around.

The name is fun, though a bit curious given that the goal is to not use any light, but perhaps Electrophorus is a bit too obscure . Single channel sequencing and semiconductor detection suggests a similarity to Ion Torrent, though presumably with Illumina's terminator chemistry (ILMN is promising similar data quality to the rest of the line, which means no homopolymer issues which implies reversible terminators).  The chemistry may well be a bit cheaper to prepare, since no fluorophores are involved.  If true, there's no guarantee that will go to lower customer costs; Illumina could just pad their margins.  Most likely, the company will pass some of the savings on, but pocket a lot of it -- but that will largely depend on the competitive landscape when the instrument launches.

Single channel sequencing implies 4 flows per position, as opposed to just one with the existing 2-channel or 4-channel SBS.  Presumably much of this time is caught up by the faster data acquisition from an electronic sensor.  The nature of the detection hasn't been broached.  Ion Torrent used the pH change from the hydrogen ion produced during polymerization, but there was talk of detecting another product of that reaction (I forget what it was).  The biggest win is eliminating expensive optics and the lasers which are both expensive and heat-generating and take a lot of space.

The library prep box is presumably going to be a streamlined version of NeoPrep, using similar microfluidic technology.  A comparison to Oxford Nanopore's VolTRAX is inevitable: both are future products (aka vaporware) and both promise to do everything.  VolTRAX is insanely small,  representing a bigger risk for potentially a ginormous gain whereas Firefly's library prep unit appears more conservative and evolutionary. Since each will be tied to their respective platform, there won't be an opportunity to mix-and-match, but in each case success could drive much greater adoption of the downstream sequencing platform.

James Hadfield has already suggested the possibility of unbundling the two Firefly components, since library prep is targeted to take 3.5 hours but sequencing might run from 3.5 hours to 13 hours.  So if your assay needs the longer, slower chemistry, one could imagine a 3:1 ratio of library boxes to sequencing boxes.  For a molecular pathology lab that wants to run multiple samples asynchronously (ala QIAGEN GeneReader), if we assume a 50:50 split in cost that means about $360K to launch a new assay every hour (please check my math!) Of course, the real math will depend on exact assay times and real estimates of lab workflows.

As far as buying one, $30K is definitely an easier sell to management than $100K or even $50K; this is a decidedly non-linear affair.  But, $30K is still a serious capital expense which requires pretty high approval at most companies, so Firefly is not getting into "we don't need no stinking management buy-in" territory.  Still, every initial purchase price decrease should steal a few more labs to buy their own sequencer, even if the new, small box is used just for method development and rush jobs while the local core or service provider handles the big jobs.  Firefly might be quite portable, perhaps enabling sequencing from a small vehicle (but probably not from a backpack). 

The bottom line, though, is Illumina is somewhat unusually announcing a technology direction well in advance of launch (though it could be argued that NeoPrep had a very long public gestation).  Perhaps this in itself shows that Illumina is trying to build a defense versus MinION and any other uber-low cost sequencing platform, making it clear to the market that they plan to play in this space by extensions of their current chemistry.  For labs already heavily invested in Illumina-related technology, by which I mean any combination of sample prep methods, instruments and downstream pipelines, that is news that can potentially keep them in the fold.  Labs which haven't already been ensnared, and even those that have been, may not have the patience to wait two years and could be tempted to try MinION (or Genapsys, should they launch soon enough).


Anonymous said...

Was it pyrophosphate you were thinking of?

Illumina sequences via in-direct means, meaning it reads the synthetic base incorporated along side the natural base. Oxford Nanopore detects the natural bases directly.

Oxford Nanopore is single molecule where amplification is not required. Illumina requires amplification.

Illumina's method cannot detect directly epigenetic bases, Oxford's can.

Interesting too that Illumina is going after cancer genomics with Grail. Pretty sure that's more difficult with their short read technology than with the long read technologies of PacBio and Oxford.

So the real question, sans spin, is what is the purpose of sequencing in the first place?

Unknown said...

Came here to say what 'Anonymous' said above - with a nucleotide incorporation, a pyrophosphate is released along with a hydronium ion, and 454 chemistry read the first by-product (using the chain reaction of DNA polymerase, ATP sulfurylase, luciferase and apyrase and two additional substrates) while Ion Torrent read the other.

Interesting implication of how a binary signal in the CMOS can determine four bases - now that's quite a trick.

Jay Flatley's comment in the Bio-IT piece said that there would be 'substantial investment needed' - they may need to set-up their own microelectronics manufacturing, similar to what Affymetrix had to do almost two decades ago.

"Anonymous" just lost credibility about his comment regarding readlengths and cancer genomics / Grail; clearly he or she hasn't read up on what Grail is about. But on the Internet nobody knows if you are a dog, so there's that.

By the way my wife and I finished 'Mr. Robot' last night, and really enjoyed it. I don't like Anonymous, because in the end, we're all people. Really.

"So the real answer, including spin, is that sequencing technology improvement is a needed component for future adoption, since it drives costs lower and adoption higher."

Thanks for the helpful writeup, Keith!


Anonymous said...

"By the way my wife and I finished 'Mr. Robot' last night, and really enjoyed it. I don't like Anonymous, because in the end, we're all people. Really."


""Anonymous" just lost credibility about his comment regarding readlengths and cancer genomics / Grail; clearly he or she hasn't read up on what Grail is about. But on the Internet nobody knows if you are a dog, so there's that."

And, what about splice variants and other rearrangements as well as long repeats/homopolymers that are common to cancer, and other diseases, that are beyond the capabilities of short reads?

""So the real answer, including spin, is that sequencing technology improvement is a needed component for future adoption, since it drives costs lower and adoption higher""

This isn't a game. What's the point in lower costs if the results aren't accurate?

Let's get to the bottom of this on scientific merit, not commercial imperatives.

Answer this question: Is Illumina sequencing really 99+% accurate?

Anonymous said...


Anonymous said...

So even though:

- the average GC content of the human genome is about 40% and Illumina technology uses PCR (Manteia invented) and PCR is well known for GC bias

- short reads can't deal with homopoylers or repeats of any significant length

- can't detect epigentic bases such as methyl, formyl, carboxy, hydroxymethyl C or hydroxymethyl T (recent paper)

Illumina technology is still 99+% accurate?

So, I guess you must also believe that the HiSeq 10x usefully sequences human genomes for less than $1000?

Keith Robison said...

Anonymous: 1st, a correction. Bridge PCR was invented by Mosaic Technologies; the IP was later bought by Manteia.

No analytical technique is perfect; all miss some features of interest and make errors. Illumina technology has proven to be very, very powerful for many types of biology; it is not able to answer every question. Methylation is an important aspect of cancer biology, but not seeing it does not preclude Illumina technology from being useful in this space, particularly for observing the recurrent driver mutations seen in cancer & a subset of which have targeted therapies.

Undoubtedly a decade from now we will have much better technology, but that won't help the patients in the nearer term.

new said...

My guess is it probably is optical. I don't think "semiconductor" and "optical" are necessary mutually exclusive? Have written up my thoughts here: http://41j.com/blog/2016/01/speculation-on-the-illumina-firefly/

Would be interested in your comments.

Anonymous said...


1) if there's one fluor label for four bases then does that mean that one base is added at a time, like 454, rather than all four labelled bases at once in the current Illumina systems?

2) can this CMOS system provide in improved results ie more accurate, longer reads?