Wednesday, October 05, 2016

ONT's Wafer Thin Update

Oxford Nanopore's Clive Brown gave an hour-long update on their platform last week.  A busy social schedule featuring (on different nights) GMO beer and challah - plus six hours travel each way to the challah-fest. Add atop that a fast-moving upper respiratory tract infection, and I'm even more behind in blogging than usual for such events (plus I gathered another post idea away -- and had another suggested to me). Time to get working on the backlog!

Clive covered a very wide range of topics, moving from immediate improvements to items not expected to be released for over a year. All but one of the announcements spring from items in his London Calling talk, so few complete surprises.  That's a bit welcome, given the possibility of grotesque explosions suggested by the event's title, "and finally, monsieur, a wafer-thin update".  I've grouped them in a way I find logical, though it doesn't strictly follow the order Clive used. 

There was some introductory material for complete novices, which I will skip over.  Brown also celebrated the success of MinION sequencing on the International Space Station, which is now available as a BioXiv pre-print.  Ironically, this zero-G success was followed by several later items that depend on gravity.  I don't remember any celebrating over closing out the Illumina lawsuit, which lifted the spectre of a prolonged, distracting and expensive litigation from the company

A suite of improvements rolling out mid-October

The announcement of most immediate impact is a series of improvements in the core MinION platform.  A new chemistry called R9.4 will be rolled out the middle of this month.  R9.4 uses a new variant on the R9 pore, apparently the result of a protein engineering campaign involving over 1200 mutants. R9.4 will operate at 450 bases per second, or a bit shy of twice as fast as the current chemistry - and Brown continues to tout the technology as capable of operating in excess of 1000 bases per second.  The new pore is partnered with a new motor protein, E8, which enables the higher speed with no loss in accuracy.  Actually, R9.4 was touted to have slightly higher mean accuracy and a tighter distribution of accuracy than the older R9 chemistry.  However, a confusing bit came up -- some of these improvements will be only available as 1D chemistry.  Apparently R9.4 will come in both a slow (250bp/s) 2D and fast (450bp/s) schemes, though it isn't clear whether this is purely driven by the library kit and running scripts or those versions will be different flowcells. Long-term, Brown sees 1D as supplanting 2D, but assured users 2D isn't going away.  ONT's field reps are probably having fun demystifying this for everyone.

ONT has sometimes briefly released protocols in which samples were bound to beads prior to loading on the flowcell.  New kits will return to this methodology -- permanently.  By loading the beads on the new SpotON-type flowcells and allowing gravity to deliver the DNA-laden beads to the pores, higher yields are obtained. Some of this may be by encouraging physical proximity, though Brown also held out the idea that the beads help keep destructive contaminants away form the pores.  

The new kits also change out the adapters. ONT's adapters have a means of keeping the pores quiet ("stalling chemistry") prior to running, but the adapters previously used were holdovers from the now defunct R7 chemistry and are now being blamed for the progressive degradation in the number of active pores during a run.  The new adapters mitigate this issue of "pore fouling".  With higher speeds, bead delivery and nearly eliminating the drop-off in performance over time (until the flowcell goes kaput, due to consumption of an electrochemical component), R9.4 is touted to deliver up to 9Gb in 5 hours.  Brown said he had targeted 10Gb, "but what's a gigabase between friends" -- and then quickly promised he hasn't quit looking for ways to squeeze that tenth gigabase out and thinks it might be achieved in "a week or two" [just before posting update - Clive has tweeted hitting 10Gb] -- which he sees as 30% of theoretical throughput with 50% his target.  

I actually missed this in the live presentation, but Clive has started talking cost effectiveness.  He's always emphasized the no capital cost requirement of MinION, but now he's talking in dollars (not pounds!) per gigabase pair.

ONT's team has squeezed elsewhere: the new rapid 1D kit requires only 5 minutes (and somehow using click chemistry, which unfortunately was a throwaway line with no explanation), half the previous 10 and also will now ship complete - no more remembering to order third party reagents.  A graphic shows the protocol having been simplified by eliminating the heating step and the need for ligase.  Brown is also bullish that a lyophilized or otherwise room temperature stable version of the 1D kit should be perfected in the near future.  Optimism was also expressed that the new rapid 1D kit will be more tolerant to contaminants in the DNA preps than the first generation kit.

A new kit in the near future will be optimized for short reads, though Brown directed any questions on this to the field support team (he actually expressed surprise anyone would want this, having apparently skipped Zev William's talk at last year's New York meeting). Additional kits will be rolled out this fall for direct RNA (to developers), rapid PCR, a new wash kit, new 12-plex PCR barcoding kit and rapid amplicon assays (not quite clear how that's different than rapid PCR).  There's also 1D and 2D ligation kits.  Sorting through this growing zoo of kits will be a task for users, and easily a point of confusion for novices (e.g. why would anyone choose the non-rapid 1D kit over the rapid kit?).  Brown also promised barcoding for all kits, though it was unclear from the graphics how or whether the rapid 1D prep could utilize barcoding.

Also part of the mid-October roll-out are new versions of the basecallers and MinKNOW running scripts to deal with the new chemistry.  The local basecaller is now the preferred method, with plans to shut down cloud-based calling (and thereby a major expense for ONT), perhaps as soon as November 5th.  However, it appears that Metrichor services (such as the WIMP metagenomics pipeline) will remain in the cloud.
One additional important change will be in the ordering process.  Previously, flowcells were delivered as ordered, making it more difficult to take advantage of bulk pricing.  In addition to the near-pore loading port, the SpotON flowcells have significant improvements in materials to greatly reduce the diffusion of air into the devices, which ultimately destroys pores during storage.  So Oxford can now store flowcells for long periods, which means they are more willing to take the risk for doing so.  Customers can commit to a large batch of flowcells, thereby triggering discounts, but then draw down that commitment as desired.  Coupled with a promised 48-hour dispatch time in the U.S. and E.U., this makes flowcell supply closer to a just-in-time delivery system.  This is also key for users since ONT only guarantees flowcell performance if they are promptly run through Platform QC. Previously if you ordered a few dozen flowcells, you were stuck QCing a few dozen flowcells on arrival.  Now that's only the case if you need to run all those at once.

In addition to the upcoming software updates, Clive also reminded everyone that the NanoNet 1D basecaller has gone open source. Nanonet allows users to train the caller on their own training sets, theoretically calling modifications or exotic artificial nucleotides. Nanonet is built atop  the OpenCL library, which supports shifting computations off to CPUs, GPUs and FPGA. 

The 2D caller and Albacore pipeline will go open source as part of the mid-October roll-out frenzy. Clive also confirmed that Oxford is actively working on Field Programmable Gate Array (FPGA) to accelerate basecalling. Clive did err in suggesting that this would represent some sort of bioinformatics first; both TimeLogic and Compugen offered FPGA-based sequence similarity search accelerators in the 1990s.  However, those were big boxes; Brown floated the idea of ultimately including the FPGA within the MinION, though initially it would be available as a plug-in board.  FPGA development was described as in "late stages".

One critical improvement not in the roll-out: fixing the homopolymer issue. The current RNN basecaller rounds any homopolymer of greater than 5 to five.  Brown expressed confidence that this can be fixed and that significant accuracy improvements can be obtained by further software tweaks, though he didn't rule out further chemistry changes to assist.  This confidence was supported with a plot showing that homopolymers beyond 5 have distinct time domain signals in the current chemistry, information that isn't fully utilized by the current basecaller. Brown curiously expressed annoyance at Ewan Birney's extremely prudent suggestion that accuracy start being reported and plotted on a log scale, increasingly valuable as the error rate drops below 90%.  Phred scores rule!

PromethION's holding pattern

Brown's update on the big PromethION boxes confirmed that the program has been held up by flowcell manufacturing issues.  Confidence was expressed that this problem with obtaining consistent yield of active pores can be resolved in "a matter of weeks." 

The instruments themselves continue to ship, though Brown also suggested that the compute system will probably be replaced early next year.  Oxford is still accepting applications for the early access program (PEAP), but that the company is considering closing the program.  No specific disclosure was made of the number of PromethIONs in the field, nor the sites that have received them, though it was suggested that flowcell deliveries would commence with ten to twelve devices in the field.  Brown also disclosed that Oxford is planning to charge $15K annually for maintenance to keep a PromethION fully upgraded.

Near-term and far-future sample/library prep improvements - and a baby flowcell

Brown updated on the full range of sample and library prep technologies which Oxford has dangled as ideas in the past.  Welcome news for both those of us who can't master the current lab protocols, as well as many of can but have more important things to do.

Registration for the VolTRAX microfluidic library prep device was announced to be opening on October 11th, with the early-access program starting around the time of Oxford's New York City meeting in early December.  VolTRAX will support a range of library prep protocols.  Initially, the user will load the reagents, but Brown envisions that in the not distant (but unspecified) future the disposable cell would come pre-loaded for a given protocol. VolTRAX is capable of both thermocycling and magnetic bead cleanups, and Brown also foresaw cell lysis moving onto VolTRAX in the future.  

Brown showed the current prototype of the Zumbador sample preparation device, which still hasn't graduated from a code name to an actual product (e.g. ZumbaDOR).  The concept is to have reagents pre-loaded in lyophilized form within the compact device as well as a number of disposable options for introducing the biological sample.  These might include a swab-like device and others that are "miniature spittoons".  Dealing with truly difficult samples, such as plant tissue, gram positive bacteria or bone wasn't mentioned.  Capillary action would move the sample through the different stages within the device, finally ending at a supply of beads which would be dropped by gravity (triggered by being hydrated) onto a SpotON flowcell's port, where gravity further directs the beads to the pores.  No projected launch date was given.

The SmidgION mini sequencer, which mounts onto a smartphone so the latter can provide the necessary compute, is now projected to launch in late 2017.  Brown promised that despite always showing it attached to an iPhone, an Android version will launch as well.  No word on whether the iPhone 7's shedding the headphone jack caused any issues for the design team. SmidgION is forecast to generate 400Mb per hour.

An interesting spin-off of SmidgION is a miniature flowcell for the MinION that apparently separates the disposable "wet" components from the pricey electronics.  Oxford had called this scheme "crumpet chips" back at the first London Calling, but that terminology wasn't used here.  The flowcells fit in an adapter (the FLONGLE) which bears the electronics, reminiscent of the adapters allowing cellphone memory cards to be used in digital SLRs (and similar).  These were originally developed to enable developing the SmidgION, but would be a welcome addition -- potentially lowering the cost of sequencing small samples for training, teaching and hobbyists.  Such flowcells (with 128 or 256 channels)  might well be popular for diagnostics, particularly as performance of the prime flowcells keeps ramping up and overshooting the requirements for amplicon-based methods.

Brown's second surprise was the announcement of a new approach to sample enrichment.  Oxford has talked in the past about approaches which would have capture oligos floating in the membranes, but the new idea (seen elsewhere, such as in several PacBio applications) is to use Cas9 to capture targets. Two versions were set forth.  In one, Cas9 with an appropriate RNA would gate DNA to the pore (I've re-watched that part of the presentation several times, and still haven't quite wrapped my head around the mechanism).  In another, a DNA tag affiliated with the Cas9 would be sequenced by the pore in a counting format.  The Cas9 technology is projected to be available in the near future.

Other far-future tech - and rumor squelching

Brown talked a little bit about possible long-range replacements for the current nanopore technology.  One approach is to replace the current ASIC electronics with ones based on Field Effect Transistors (FETs); this would allow potentially 1 million channels in a MinION flowcell (as opposed to 512 today) and a 3 minute human genome.  Brown estimated this is at least two years away.  Oxford also reported progress on solid-state nanopores, generating similar electrical signals to the current biological pores.  Brown envisions this system retaining protein-based driving of DNA through the pore, but would still gain significantly in environmental hardiness

In the Q&A session after the presentation, Brown swatted down the persistent rumor of a MinION MkII (which I'm a source for; my London Calling 2015 report, based on tweets as I wasn't there, describe one).  MinION will remain at 512 channels for the forseeable future (barring the FET technology described above), with performance improvements coming from faster pores and keeping those pores working as much as possible. 

Closing Thoughts

MinION steams ahead, and R9.4 has already been getting rave reviews on Twitter by the advance squad who has used them. Oxford has generally disdained gunning for PacBio, but an increasing case can be made that MinION can substitute for PacBio for some applications.  The big catch there is either the application must just not care about not getting homopolymers right or Illumina data needs to be provide to clean those up.  PacBio has had its own delays getting the new Sequel devices off the ground; company data looks quite striking in terms of quantity, but word on the tweet is that sites haven't yet gotten it.  The mini-flowcell pulled from the SmidgION project could be a huge addition to the lineup, particularly if priced very aggressively so as to serve as a gateway drug for greater use.

PromethION is clearly having a difficult birth.  As with other platforms that viewed scaling an electronics-based system as easy (e.g. Ion Torrent), scaling is hard.  PromethION uses a different flowcell form factor with different electronics and higher density, meaning it is almost a completely different entity. Certainly lessons learned on MinION can be transferred over and all the kits are the same, but PromethION is going to tie up a lot of developmental resources.  Conversely, once the instrument starts delivering it will raise the real spectre of over-stressing the downstream computational processing capabilities.  This is illustrated by the planned compute upgrade, and the FPGA basecaller push.

The new ordering scheme is welcome, allowing even low frequency customers to get ordering discounts yet receive flowcells as they need them.  Oxford's efforts to eliminate the need to refrigerate components will have a huge impact on field use. They will also help in the lab, since in the current scheme parts of the MinION operation must be stored in three different places (ambient, 4C and -20C), making it easier to lose track of valuable equipment and reagents in the chaos of a multi-person lab.

Most of the rest are the sort of dream pieces that keep us all excited, but are too far distant to really plan on.  This is particularly the case with PromethION struggling; pressure to shift resources to this flagship product must be frequently intense.  VolTRAX's beta program should launch this winter; hopefully Oxford will have incorporated lessons from the bumpy launches of MinION and PromethION.  Zumbador and SmidgION are exciting, but very distant, and the possible FET or solid state pores probably more so.

On the wet side, and even faster 1D prep is good, unless those 5 minutes are too complex a choreography of actions.  However, if it is then that will drive VolTRAX in this application; if not many may choose to stay manual. The short read kit will be interesting for those developing copy number and other counting applications.  The utility of the Cas9 schemes really depends on their performance, as Cas9 isn't quite magic velcro.  The cost of synthesizing the required RNA probes for Cas9 systems can't be discounted either.

Oxford, as always, has a lot of exciting ideas.  But as the company matures, execution of the ideas will be increasingly weighted in evaluating the company.  Oxford must also find ways to continue to grow its following from a small cadre of committed loyalists to broad swaths of professional biologists.


Mike Schatz said...

Nice writeup! Although I think you have your throughput numbers off on the MinION - it can reach up to 10GB in 48 hours, not 5 hours. Here is Clive's plot about exactly this from yesterday:

Keith Robison said...

You are indubitably correct; need to check what I hear against axes!

wlad said...

No news about their DNA synthesis spinout?

Keith Robison said...

wlad: no, nothing on the synthesis bit. Only comment on sequencing proteins is that it is far off.

Clive G. Brown said...

Our synthetic DNA/biology device business is a subsidiary and will have its own previews/announcements/updates.

Anonymous said...

FETs "at least two years away" is such a gross exaggeration even by the stretchy standards of ONT vaporware. They are usually 2-3 years late, including with most of what they promised at London Calling 2015. Then again 10 years falls into the "at least two years" window so technically this is not a white lie. Technically...