I want to be very clear on a few points before I dive into some criticisms. Pacific Biosciences has an amazing technology, just short of sorcery. They watch individual DNA polymerases operating in real time; just savor that concept for a moment. This is accomplished via an astounding physics trick that enables illuminating a vanishingly small volume containing the polymerase without illuminating the mob of fluorophore-tagged nucleotides swimming behind it.
PacBio taught the world how to use long, noisy reads, at a time when many thought that this was sheer folly. Innovative scientists at PacBio such as Jason Chin (@infoecho) have developed amazing software to use these long reads to solve increasingly challenging projects, starting with hybrid assembly (using Illumina data to clean up the PacBio reads), then straight PacBio assembly of haploid genomes and now assembling diploid genomes with PacBio reads. They've collaborated with other companies, such as BioNano Genomics, to assemble a complete human genome de novo and a fish genome as well, achieving amazing quality. All that amazing, sophisticated software? PacBio has made it standard policy to release all of it open source.
PacBio taught the world how to use long, noisy reads, at a time when many thought that this was sheer folly. Innovative scientists at PacBio such as Jason Chin (@infoecho) have developed amazing software to use these long reads to solve increasingly challenging projects, starting with hybrid assembly (using Illumina data to clean up the PacBio reads), then straight PacBio assembly of haploid genomes and now assembling diploid genomes with PacBio reads. They've collaborated with other companies, such as BioNano Genomics, to assemble a complete human genome de novo and a fish genome as well, achieving amazing quality. All that amazing, sophisticated software? PacBio has made it standard policy to release all of it open source.
Personally, they've always been the nicest of people, no matter what I've written about the company. In my professional work, PacBio non-hybrid long read assembly was truly a quantum leap, changing a critical problem from one I could not solve except with by a lot of other scientists' lab work and a ginormous dose of luck, to one we solved routinely. PacBio scientists were generous with their time to help me get this up-and-running. I can't say enough nice things about the company and its personnel.
Unfortunately, if I turn off my personal warmth and look with a critical eye, PacBio is a threatened company. There are three general classes of existential threats to the company, and unfortunately the marketplace cares not about them being a good citizen or a pioneer. Should better technology come along, PacBio is in trouble.
Both the glory and the Achilles Heel of PacBio are their instruments. As extolled above, they are wonders of technology. They are also very complex optical devices with moving parts to allow hands-off operation through multiple flowcell runs. The Sequel improves substantially on the the original RSII, being about half the size and weight and only $350K , but it is still a large, expensive box in a world which is moving away from that.
PacBio has several other vulnerabilities. For good results, a very labor-intensive library prep is required. Ultra-long reads (>50Kb) are still relatively rare; laser-induced polymerase damage still leaves an exponential decay component to the read length distribution. They've worked wonders showing 85+% accuracy reads are usable, but there is still market resistance to that level of accuracy.
The first threat to PacBio is Genia, which may well deliver very long reads with very high accuracy on very compact instruments. Genia hasn't yet launched, and it isn't clear how close they are, but should they do so it could be very devilish for PacBio. Awkwardly, Genia was purchased by Roche, which is PacBio's key strategic partner. Perhaps they have a tacit agreement to carve up sequencing applications space, with Genia going after markets with poor penetration or applicability of PacBio (exomes and targeted panels). But it still can't be very comfortable watching your partner incubate a potential downfall.
The second threat are linked read sequencing chemistries such as 10X Genomics and iGenomX. These enable gaining long-range information on Illumina hardware. Both require some level of expensive equipment (about $150K for 10X and $250K for iGenomX, but the tradeoff for a big lab is that they can shift their library-making onto these platforms. Linked reads can capture very long range information (>100Kb) that is difficult to get on PacBio, but conversely these short read platforms can't be expected to resolve long VNTR repeats (such as the CAG repeats in Huntington's disease), and may have difficulty resolving very long, high accuracy repeats (this really hasn't been properly explored, IMHO). These systems can perform with low nanogram amounts of input DNA; PacBio appears to require at least 50 nanograms of input material. So it is a complex decision as to which platform to use, with not all the data one would like yet available.
Oxford Nanopore offers a particularly devastating scenario for PacBio. PacBio's one big advantage over MinION is the error profile; PacBio has purely random errors, whereas MinION has a significant systematic (primarily homopolymer) error problem. Otherwise, MinION has a lot of advantages. Oxford essentially gives the devices away, requiring no capital cost. The newest chemistry is delivering 5-10Gb of data per run. The newest (but not yet available) library kit will require only 5 minutes of hands on time, which is also the amount of wallclock time. This will replace the extant 10 minute rapid kit.
With my previous diving into the Illumina-Oxford dispute, I've (for better or worse) established myself as someone willing to do this again.
@OmicsOmicsBlog Care to do some patent digging? https://t.co/gJc9zRKsoV
— Lex Nederbragt (@lexnederbragt) November 2, 2016Part of my ethos is that if charged with a task, I dive in with unbridled enthusiasm
— Keith Robison (@OmicsOmicsBlog) November 2, 2016
On the bright side, this meant that some initial explorations were done before I even knew the trade action had been initiated.
@BioMickWatson sounds like it. US009404146 claims a lot @OmicsOmicsBlog @lexnederbragt pic.twitter.com/71fZ43Q4p9
— James Hadfield (@CIgenomics) November 2, 2016— Lex Nederbragt (@lexnederbragt) November 2, 2016Okay, I've skimmed through the patent twice. PacBio definitely thought ahead when they wrote their patent for Circular Consensus Sequencing. As written, it covers a wide range of schemes (including nanopore sequencing) in which a hairpin adapter enables both strands of a template to be read and the information combined to yield a more accurate consensus. One quirk that may be due to prior art: explicit reference is made to inserts of 500 bases or more.
So Oxford has two possible routes forward. The less likely one, in my opinion, will be to vigorously attack the PacBio patent. I haven't found any devastating prior art, but I've also made a cursory look. ONT will obviously be far more motivated. Or perhaps they would claim the idea of combining reads from both strands was obvious, given the routine use of this approach in Sanger sequencing. But Sanger sequencing isn't single molecule sequencing and two independent reads from a cloned fragment is different than running around a hairpin. Obviousness rules aren't remotely obvious, so this is a point of uncertainty.
But there is an obvious course forward for Oxford: abandon their double-stranded (2D) sequencing kits, at least in the U.S. (I haven't found an international companion to the US patent). Oxford's Clive Brown recently expressed a preference for a purely 1D platform being the future of Oxford. If PacBio had been paying attention, the base accuracy for 1D reads is around 90% now, better than PacBio. Except, of course, for the systematic homopolymer problem, but it is dangerous bet to think that Oxford can't ameliorate this with better software or further engineering on the pore and motor proteins.
@BioMickWatson @OmicsOmicsBlog @lexnederbragt I suppose this is another reason why ONT could prefer people to use their 1D protocol— David Eccles (@gringene_bio) November 2, 2016
So PacBio has filed litigation that Oxford can again play matador to, yanking away the cape to expose a void. At least with Illumina this wasn't obvious at the time the litigation was filed; Oxford must have been in the licensing discussions for the CsgG pore already, but was able to dramatically reveal it. This time Oxford's feint option is already known. Tilting at this windmill can't help but appear to be a desperation move and one which could erode the support PacBio has gotten from stock analysts. I welcome input in the comments, but winning a trivial tactical victory makes no strategic sense to me.
20 comments:
Hi Keith, thanks for the interesting post.
You have a very good summary of the threats to Pacbio.
Personally, I wouldn't really consider 10x or IGenomX as much as a threat if for instance, Pacbio reduces their cost as they hope to by loading more ZMWs. And these technologies would perform better on Pacbio over Illumina data.
Regarding nanopore technologies, you mention the usual systematic homopolymer error, has there been any public research suggesting that this can improve ? This is a problem that all nanopore based techs have and have not solved for a while. The only direction appears to be searching for new pores and that appears to have only swapped one systematic error for another unless i am mistaken.
I am curious to know what makes you optimistic that they can reduce this error, that would be very interesting. It's a long time waiting, much (?) longer than PacBio took to catch up with its earlier promises esp considering the rewards up for grabs.
Also in terms of just using 1D, isn't that going to make it in general less reliable over using 2D ? I guess CCS is definitely a special feature for PacBio, enabling the same base to be read etc so it's not surprising they are defending it.
One thing that PacBio has in their favor is the number of papers that are being published using their technology. That's an early indication of a good technology.
In a field that relies on openness and transparency, I guess ONT's restrictions on who can get their pilot minions do not reflect as well.
For instance, ONT announced on 20th October that they had sequenced the human genome on a minion, but no data was released and still hasn't been released to my knowledge... It appears to fit a pattern of a lack of transparency atleast for a novice observer like me.
While one would expect nanopore tech to be eventually successful, one would be equally worried if they put all their money into Oxford Nanopore i think.
Thanks !
Mohan
Adding to what Mohan has said,
Actually, PacBio have been on the market for maybe 8 years now and their accuracy and throughput have not improved much at all as if they have hit some sort of limit imposed by the measurement method. ONTs rate of improvement seems much faster in just 2 years, and i saw some videos online where they say the homopolymers can be dealt with in consensus initially and then in single molecule. I understand there are public domain tools that already do the consensus corrections ?
Watching ONTs presentation the 1 D seems more reliable and easier if i understand it with, as keith says, an error rate better than PacBio.
You can't claim PacBio have 'early indications' of a good technology, its over 8 years old, with ONT being 2 years. As I recall for the first 18 months or so of RSII outage there was very little if any data and no 3rd party publications at all.
I think ONT only restrict direct competitors from MinION purchase, they say its not a pilot but is fully commercial with 2-3000 out there.
As for their human genome, I think they said they'd done a preliminary analysis and would make it, or an updated version of it, available once they'd been thru it. It must take a fair bit of effort to scrutinise your first Human Genome, no doubt detractors would be looking to jump on, and exploit, any flaws or oversights - am i right ?
Interesting comments Mohan, but they don't tally with what I've been able to find in the public domain.
Dicky
Is this comment from Mohan Holenarasipura, apparently a PacBio employee ?
Probably not a comment from the same person, more like racist innuendo, considering that Mohan is an extremely common South Asian name...
I think the question is more about the affiliation of the commenter with the legal aggressor, and the apparent bias in the comments, than a race thing surely ?; and how many Mohans can be working for the handful of DNA sequencing companies? If the commenter had signed off Felicity, and theres one Felicity on LinkedIN working for PacBio, would it be unreasonable to ask if its the same Felicity ? or would it be sexist ?
Hi,
No I think you're wrong i think its Mohan from UniCredit, an investment bank in London. Are UniCredit a big PacBio investor ? if so, then they are down 20% today and possibly a lot more to lose - what a waste of money. Perhaps that explains the "put all your money" comment ? I dunno.
Felicity [not PacBio]
Hi,
Just to be clear, I have no affiliations with any of the companies, pacbio, oxford nanopore or Illumina.
I was, as you can see from my comment above, biased towards Pacbio based on what I read, their papers in the big journals etc, and wanted to find out more considering their stock is now much cheaper to invest in. But it appears the comment section of this blog is not the place to find them.
Thanks anyway.
Mohan: For the record I really think you ask some good questions - please be a little patient with me for a full follow-up & ignorethe silliness above.
Hi Keith, really appreciate that. Thanks for writing back, I shall wait for your answer.
In response to Dicky
1) Pacbio hasn't been on the market for 8 years. They've been on the market only 5 years. But both ONT and Pacbio were founded in 2015.
2) Pacbio went from 1k mean RL and 75k ZMW's to 20k mean RL and 150k ZMW's in 4 years. That's a 40x improvement in throughput. ONT has also been rapidly improving their chem. But a lot of the large throughput numbers are just very long run times >24h. You could take an RS chip and probably do some type of rinse and reuse to also get super long runs with high throughput.
3) ONT is still a bit off from being commercially ready. Hand picking a few data points is easy, but having a product that works reliably for thousands of customers will be a big challenge. It took Sequel 1 year to really start performing how it should. It will take ONT a while to adjust from a development product to a commercial product.
I don't think this lawsuit by Pacbio was done is desperation. It's very likely ONT is infringing on a patent. And it would be stupid for them not to go after it. Don't forget this is corperate America, where companies like Apple will sue over swiping.
Also being forced to only use 1D reads will be a decent blow to ONT. Even if a 1D ready is in the 90-95% accuracy it still will be lower than the 99%+ by a 2D read.
As to ONT claiming that they are commercial, not sure how anyone can objectively believe that. Until you can order one without restrictions they will remain in only what can be described as an early access program. Just because they conveniently drop the "early" part of the name does not make it commercial. ONT can go and order a Sequel. ONT will be commercial when PacBio can order a Minion.
As to PacBio lacking improvements - they went from 100k ZMWs to 1M on the Sequel. Their read lengths and output per flowcell have been going up steadily. And they have a hard working ethos to them, not some jerk CTO that makes comments unbecoming a corporate officer, e.g. accusing scientific competitors of plagiarism without a shed of evidence.
One thing that must be noted is that because of the undisputed breakneck rate of improvement in their performance, both PacBio and ONT are rapidly coming up against physical limits of what the measurement technology is capable of. The next wave of improvements will have to come from the biochemistry. And everyone knows (or should know) that any hope for exponential improvements in chemistry is a pipe dream. Our phones would not be stuck at 12 hrs of battery life under moderate usage if that was not the case.
As to this comment :
"As to ONT claiming that they are commercial, not sure how anyone can objectively believe that. Until you can order one without restrictions they will remain in only what can be described as an early access program. Just because they conveniently drop the "early" part of the name does not make it commercial. ONT can go and order a Sequel. ONT will be commercial when PacBio can order a Minion. "
Anybody can order a MinION, except a direct competitor - why would a direct competitor order one?. You are wrong - NOT everyone can buy a Sequel because first you have to rustle up 350k dollars. How many post-docs (or anybody) can do that ? Further, whole countries and their relatively poor scientists are excluded from buying large boxes given the capital requirements. So what, pacBio can't buy a MinION, an Ebola researcher in Guinea can !
This "its not commercial" meme, given the numbers out there, is pure downplay. Seeking to belittle the threat and maintain a chauvinistic position.
"One thing that must be noted is that because of the undisputed breakneck rate of improvement in their performance, both PacBio and ONT are rapidly coming up against physical limits of what the measurement technology is capable of. The next wave of improvements will have to come from the biochemistry. And everyone knows (or should know) that any hope for exponential improvements in chemistry is a pipe dream. Our phones would not be stuck at 12 hrs of battery life under moderate usage if that was not the case."
As I understand the technology this statement is utterly false. The difference between PacB and ONT is PacB currently rely on advances in third part camera sensor technologies which may not be incentivised to move in their direction. ONT designs and makes its own sensors.
Some interesting and salient points in here - I've commented beneath each :
In response to Dicky
1) Pacbio hasn't been on the market for 8 years. They've been on the market only 5 years. But both ONT and Pacbio were founded in 2015.
>> I think you mean 2005 ? Actually PacBio claims it was founded much earlier. The RSII went into first shipment in 2010, so its six years old versus ONTs 2.3 years.
2) Pacbio went from 1k mean RL and 75k ZMW's to 20k mean RL and 150k ZMW's in 4 years. That's a 40x improvement in throughput. ONT has also been rapidly improving their chem. But a lot of the large throughput numbers are just very long run times >24h. You could take an RS chip and probably do some type of rinse and reuse to also get super long runs with high throughput.
>> This is utterly incorrect, their run time has been 48hrs since the MinION came out. ALL of their throughput improvement comes from the speed of translocation of the DNA coupled with having more of the available channels occupied by single pores.
3) ONT is still a bit off from being commercially ready. Hand picking a few data points is easy, but having a product that works reliably for thousands of customers will be a big challenge. It took Sequel 1 year to really start performing how it should. It will take ONT a while to adjust from a development product to a commercial product.
>> They have 3000 machines out, most or all paid. Id say that was commercial. Sure early technologies have variability.
I don't think this lawsuit by Pacbio was done is desperation. It's very likely ONT is infringing on a patent. And it would be stupid for them not to go after it. Don't forget this is corperate America, where companies like Apple will sue over swiping.
>> The infringement arguments will come out. It is also useful to use the ITC to kick potential competition.
Also being forced to only use 1D reads will be a decent blow to ONT. Even if a 1D ready is in the 90-95% accuracy it still will be lower than the 99%+ by a 2D read.
>> True, although throughput doubles with 1D over 2D, even so, accuracy still better than PacBio. They do need to fix the consensus calling of homopolymers.
The minion is not a commercial product, clearly. Actually, go look at ONTs latest accounts (available to end the of 2015 on companies house). They recognized 750KGBP in revenue for sales, against a cost of goods of 500KGBP. That's a horribly low markup (Illumina markup is something like x10 cost of goods). But even so, it probably doesn't reflect the true cost. There a whole bunch of places they good be hiding costs just to make it look like the minion is sold at a profit in the accounts (e.g. is the cloud compute included, or is that a "research expense"?).
Anyway... as always timing in important. I'd guess PacBio are doing this now because ONT are looking to raise another round in December. That's also probably why ONT are doing their "user conference" now, to increase investor confidence prior to the raise.
They'll say anything to raise another round (which they need, check their accounts). Back in 2012 they said they had a 3% error rate (obviously when shipping devices appeared that was not the case). There's probably some truth here but a lot of these statements are driven by what they think their investors want to hear.
I've still not even seen 90% overall accuracy rate properly substantiated. I've seen graphs that show that there's a population of reads, some of which have an accuracy of ~90%. But I've not seen anyone take a simple dataset, align it, and then report an overall error rate. If there's a public dataset that does that it would be great to see.
Unless I can see that, I'm going to assume accuracy is still around an 80% average (which sucks).
"I've still not even seen 90% overall accuracy rate properly substantiated. I've seen graphs that show that there's a population of reads, some of which have an accuracy of ~90%. "
Distributions of error for all reads, aligned as you say, have been widely published and the data made fully available, they clearly show ~87-93% error for 1D reads and ~97% mean error for all '2D' reads. You could of course buy a MinION and find out for yourself within about an hour of prepping a sample (unless of course you work for PacB so cannot by a MinION).
"That's also probably why ONT are doing their "user conference" now, to increase investor confidence prior to the raise."
As i Understand they do two user conferences per year, one in London in May (for the past two years) and one in New York in late November. There have been four so far.
I'm not a scientist just an investor, but as far as I've read through here, everyone has conflicting information so it's all useless info. Talk about alternative facts! I would expect more from the science community.
Better get your patent hat back on:
http://www.lifesciencesipreview.com/news/pacific-biosciences-sues-uk-company-over-dna-sequencing-technology-2082
I don't profess to be knowledgeable about this area but I don't see how it relates to PacBio's tech though? I hate to say it but it smacks of them just being patent trolls, which is a real shame.
Who knows, in their defence I saw Alexander Wittenberg's tweet about the Sequel only managing 5-6 GB and then Nick Loman's tweet almost the next day of a 5 GB Minion run that had 800kb reads in it. That's hard for PacBio to compete with on a science level. Couple that with Roche dumping them then maybe it is a lot worse at PacBio than we all thought?
Post a Comment