Wednesday, May 29, 2019

ONT's London Calling Announcements, Pass One

The annual big Clive Brown London Calling talk had a twist this year  Brown introduced a set of senior lieutenants at Oxford Nanopore and then tacked on a few "and one more thing" moments at the end.  If you're looking for the TL;DR version from me, it is that there were a lot of sensible but non-Earth shattering updates from the team followed by Clive updating on one prior crazy idea and throwing out two more -- one of which I love and the other I believe should be thrown out posthaste.
Given that nanopore sequencing is entering its fifth year in the wild soon, it isn't surprising that a lot of company updates are taking care of immediate customer needs rather than projecting far in the future. As heard at the conference, nanopore sequencing is certainly not a sideshow but a serious endeavor, driving precision medicine and contributing to the identification of illegal shark harvesting and so many more solid applications.  Oxford's platform has many areas for improvement -- which any system will have since not every issue can simply be thought out in advance.

Both in the evening talks and in the morning "Data for Breakfast" session, one area of improvement is simply the MinKNOW control software.  For example, as the volume of data coming off the system has grown, the practice of writing each sequencing read to its own FAST5 file has become problematic.  So now MinKNOW supports FAST5 files holding multiple sequences and ONT has a toolkit to build tools to manipulate these collections.  MinKNOW is getting more updates: compression, modified base calling built in, 

Lakmal Jayasinghe covered the R10 pore, thanking the community for patience in the slow rollout of the R10 flowcells.  He cited multiple reasons that ONT was unhappy with their earlier attempts at producing the flowcell: the pores were not as stable, the yields were lower and the overall signal was weaker.  R10 pores have a longer neck and two areas of close interaction, so-called "reading heads", which greatly improves the differentiation of long homopolymers.  A quick way to think about it is that the R9.x pores cover about 5 nucleotides, so expecting them to reliably distinguish homopolymers longer than that wasn't very realistic.  

Jayasinghe described in modest terms  some very unmodest efforts at protein engineering.  To improve the signal amplitude over 500 mutants of the pore were examined, with an unspecified number of additional mutants explored to improve the ability of the pore to capture adapted DNA molecules.  With these and corresponding software improvements, ONT now believes they can reach phred scores in the mid 40s with R10 alone rather than mixing R10 and R9 data -- but they have been working on versions of Medaka that can leverage both.  Jayasinghe also talked of further variants of the R10 pore with different signal properties, called R10b.  Nanopore is also exploring yet another pore called R11 which is again long with two reading heads, but very different from R10 or R10b.  But most importantly, R10 will ship next month for MinION, mid-summer for PromethION and early fall for Flongle, and users can alter existing shipments to change over to the new cells.  Also important is that R9.4 isn't going away; users can order one or the other or a mix. 

Stuart Reid had previously gone over some of the improvements to accuracy that have come from the flip-flop basecaller, which is now available to all MinKNOW users.  The direct RNA caller has been upgraded so that now its modal accuracy is 95%, just like the DNA caller.  DNA calling in MinKNOW will soon have options for calling E.coli-style (Dam & Dcm) methylation and human-style (CpG) methylation.  Homopolymers will soon be modeled using Run Length Encoding; the length of a run will be represented by a distribution rather than a fixed value.  

Reid as reviewed ONT's Medaka polishing tool, which is now outperforming Nanopolish and Racon in their hands -- and Medaka uses FASTQ/FASTA not signal data.  Further improvements will reduce the dependency of Medaka on initial assembly quality and take advantage of covariate information from the basecaller.  A pre-assembly corrector is also promised. With R9.4 data, Medaka on one bacterial genome hit Q44, though on others it was just under Q40.  

Jayasinghe noted that ONT still has multiple "horses in the race" for single molecule accuracy -- linear consensus, circular consensus,  rolling circle amplification. Unique Molecular Identifiers, chemical modifications. It appears that 1D^2 is on life-support.  ONT promised to continue supporting the R9.5 flowcells for it on MinION/GridION and PromethION but will not supply such cells for Flongle.

An interesting issue is that the accuracy of later copies of a data drops if those copies have a reverse complement already through the pore, which ONT reasonably interprets as being due to the DNA reannealing on the trans side of the pore and altering translocation kinetics.  This was also seen with the old 2D chemistry.  Chemistries based on repeating the original molecule tend to provide both forward and reverse copies and be subject to this.  ONT and others also believe that such reannealing on the trans side is behind blocking that can only be reversed by nuclease treatment; such secondary structure prevents ejection of the strand.  ONT favors their linear consensus proposal to be best because the original template sequence remains distinctive, enabling modification reading.  They are also still working on the "8b4" chemistry in which modified bases are doped in to break up homopolymers. 

Andy Heron described efforts to improve chemistry.  The Direct RNA will be getting a motor which is twice as fast, raising output.  But it will still be much slower than DNA.  There was essentially no discussion of moving to 1000 bases per second for DNA.  

Nanopore is going after the problem of translocation speed dropping across a run.  This is due to ATP consumption, as any helicase motor in the flowcell will consume ATP even if it is not driving a DNA through the pore.  Hence, high pore occupancy does not ensure maximal output without intervention.  The current solution is to pause the run (now better supported in MinKNOW), add ATP, and restart the run.  MinKNOW now reports the translocation speed, which can assist in planning this intervention.  But ONT will be rolling out in September an ATP regeneration system. Long-term, they hope to engineer the helicases to not burn ATP if idle.

VolTRAX 2 is now a product, with a genome sequencing via transposition cartridge.  The system is enabled for PCR.  This summer a whole genome amplification and the genome multiplexing kits are planned for relase and this fall 16S, rapid PCR barcoding and cDNA PCR.  Ultimately direct RNA and sample extraction will be available, as will user programmability of the system. 

For devices, PromethION with 48 cells is already shipping next month and will be joined in July by the 24-cell version.  MinION Mk1C is scheduled for August.  MinIT owners can get a trade-in credit and ONT is also offering a credit for returned flowcells which can be applied to Mk1C.  "We believe your Mk1B belongs in a drawer".  

ONT says they had a lot of pushback on the $5K starter pack for Flongle with 48 flowcells and one adapter., with the complaint being the cost-of-entry was too high.  In response, a new $1860 starter pack option has been added, with one adapter and 12 flowcells.  Individual adapters can be purchased for $750 and single flowcells for $90.  But remember: none of those include shipping and handling, which can be substantial.  So those single flowcells will be useful once you are full-up on adapters, but I'd recommend you don't start a Flongle-a-week habit.

Clive talked about both Ubik sample collection device and the SmidgION cell phone attachment; he doesn't want to release SmidgION until a sequencing mode requiring no laboratory gear is enabled.  This would be Ubik.  The final live demonstration showed off Ubik, which has evolved since last year. The workflow is more complicated now.  It's still spit-in-a-tube and shake large metallic beads inside to lyse and mix, but the tube is now a standard microfuge tube and not some kit-bashed gadget.  The exothermic reaction from adding water to  calcium chloride is used to perform the heat-kill of transposase.  There are some other liquid transfers, but with simply bulbs and no user requirement to measure.  So not the spit-shake-drip into sequencer concept Clive had hoped for, but something which requires little skill to execute.  

Clive Brown's final bit covered four "skunkworks" projects.  One is the effort announced last year to move sequencing onto the VolTRAX platform; that is progressing and he hopes to show it off next year.  Squiggles (raw signal) for three droplets were in a corner of the slide.  This application demands both changes to MinKNOW and the VolTRAX cartridge (to reduce noise).  

His last announcement was on a concept for DNA marking (not writing!) and reading for digital information storage  that I will cover (to be honest, cover means try to bury) in a separate post.  In between was an exciting new development and something enabled by it.

A major cost item in the Nanopore system is the Application Specific Integrated Circuit or ASIC, which takes the raw electrical signal from the pore electrodes and digitizes it.  MinION and PromethION flowcells have the ASIC in the cell and the ASIC is expensive, so flowcells are not disposable (ONT tries to recycle them upon return, at least for use in their R&D group).  Flongle flowcells shed their ASIC, moving it to the adapter.  But this required a fancy (and potentially point-of-failure) contact to transfer information from the flowcell to the adapter. 

What ONT is rolling out is a new ASIC  with around 400 channels which is much smaller (able to fit inside the crown on a UK penny) and cheaper than previous ASICS -- so now it can be disposable.  SmidgION's flowcell will use this.  But more exciting is Plongle, a 96-well sequencing device which would have a flowcell in each well with about 70 channels per well and targeting 0.5 to 1 gigabase per well.  That would be plenty for de novo sequencing of bacteria.  Or for multiplex PCR amplicons. Or numerous other applications.   I've discussed this sort of concept in at least my pre-meeting post; I've wanted something like this for a while.   It wasn't clear if the flowcells (or, as ONT labeled them, flow wells) would be fully independent of each other -- could you run half a plate today and half tomorrow?  ONT is targeting this for launch in the second half of next year.

One omission from the slides this year was discussion of solid state nanopores.  In response to a question, Clive downplayed their value, saying that protein nanopores are too often underestimated and the power of mutagenesis to tune performance is a huge leg up.

I hope  that covers everything other than my noted deliberate omission.  In a pair of posts soon I will gush more about Plongle and balance that (and perhaps then some) with my coverage of the DNA marking idea.

[30-05-2019 -- two small corrections. R10 has already passed Q50, not Q40, at least on Staph aureus.  P48 was actually shipping before the conference]

No comments: