Tuesday, May 21, 2019

Care to Pore Over My Laundry List?

London Calling is this week, so get ready for lots of Oxford Nanopore in my Twitter channel and if I'm on the ball, in this space.  ONT has made a number of releases and updates and there have been other developments within the nanopore ecosystem which I've failed to report.  Here, in not terribly organized fashion, is a laundry list of things I'm thinking about going into the meeting

Could Someone Loan Me a Time-Turner?

The meeting has expanded to a full two-and-a-half days, starting Wednesday around lunchtime.  I felt schedule-crimped into not arriving a full day early, which I'm now regretting -- with the overnight flight I'll start out tired.  On the plus side, apparently I get my first flight on a monster Airbus 380.  But even more daunting are the simultaneous sessions -- five going on at once and I'd almost always want to strongly be at three of them (not that I'd be bored in any of them.  I'm hoping folks tweet heavily from the 80% I can't sit in!

Flingle Me Some Flongles!

Oxford released the down-sized Flongle system in earnest this year.  Unfortunately, the initial purchase arrangement is one adapter and fifty flowcells for around $5K.  In addition to a voucher for the field sequencing kit (another recent release -- room temp stable library reagents), delegates have the option of buying a special pack of one adapter and four flowcells for $400.  

I believe -- and some colleagues agreed today -- that there's a huge market for running many of these in parallel.  If your problems are Flongle-sized, then outfitting a GridION with five of them would make total sense.  In a classroom, one Flongle per student pair should be a target -- if not just one Flongle per student.  

It wouldn't surprise me if there is a market for much higher multiplicity Flongle devices.  GridION's compute should have no trouble keeping up with 5 Flongles if it can keep up with 5 MinION flowcells -- so perhaps it could run 10 or 20 Flongles at once.  Maybe a big grid, but perhaps a radial set of sequencers arranged like petals would be more aesthetic?   Or a big carousel of Flongles ala the old Kodak slide projectors?

Some folks might not be interested in Flongles or just not interested enough to spend $400 on one -- will a secondary market emerge for adapters?  Don't quite have official bankrolling of such an effort, but no reasonable offers will be ignored out-of-hand!

Accuracy Creeps Forward

Base calling errors remain a challenge for some applications of long nanopore reads; I really don't like going to our designers and saying "if you'd just get rid of those damn 6xA sequences in your designs I could sequence this -- well, and also impute some bases since we know the reading frame is preserved".

R10 flowcells are now in beta -- I haven't gotten my paws on one which is frustrating.  ONT also released a two-choice version of their basecallers -- a slightly faster mode and a slower but more accurate mode.  

At the San Francisco meeting ONT mentioned a slew of different approaches to address accuracy, some intended for niche applications and other broader.  These included incorporating base analogs, unique molecular identifiers, rolling circle amplification to generate repeated templates, improvements to 1D^2 and probably some others I'm forgetting right now.  It will be interesting to see which have updates, which are mentioned and which seem to have slid below the waves.

Also interesting will be to see what ONT or academics are doing for combining R9 and R10 data.  They're supposed to have somewhat different error profiles, so the best polishing would put the two together in some non-trivial way.  At a minimum, the two would be marked as different.  By the way, I'm impressed with what ONT's Medaka polishing tool can do using just FASTA files, though I've seen it spit the bit on hexa-A.

HMW Purification & The Long Reads Club

At AGBT some of the nanopore mafiosos launched a Long Reads Club (complete with T-shirts!) to focus on DNA purification methods for generating whales.  Somewhere this winter the Circulomics Short Read Eliminator (SRE) started generating a lot of buzz online, knocking out much of the chaff below 10kb.  Some folks are now subbing in SRE for all the bead steps in the LSK109 ligation protocol.  In the other direction are folks now trying out non-bead old school purifications with sodium acetate precipitations.  And somewhere I have a bookmark to a tweet on other companies trying to launch kits in the long DNA prep market.  

Clive's Follies?

Okay, I'll confess that I become less-and-less enthused about SmidgION and Ubiq -- the former a smartphone adapter flowcells scheme and the latter the spit-and-sequence library prep scheme.  I fear that development effort on these is starving other efforts I care more about.  VolTRAX is another gadget that might fall into that category -- still unclear if there has been significant market uptake.  Of course, when I see them in person I find them fascinating.

The idea of sequencing on VolTRAX is intriguing, but it's another hard problem with a cool solution that's looking for a major application.  Clive suggested single cell, which would be neat -- but also pile on a bunch more difficulties.

On the other hand, I can see the use of really parallel arrays of devices with even less capacity than a Flongle.  Perhaps P1000 tips could be the form factor for a PipettION for high-throughput screening.  

Should MinIT Be Replaced by MinHOUR?

The MinIT and MinION Mk1C aim to solve the problem of getting your MinION off the ground -- built-in informatics support for basecalling and later applications.  MinIT is just the informatics whereas Mk1C is a fully integrated sequencing device.  Great concepts, but already there are reports that really good flowcell runs overflow the on-board storage capacity of the MinIT. Ditching FAST5 format is an option -- if you don't want to be able to re-call data with the ever changing basecallers or you care not about modified base calling.  It's a case of ONT being bitten by their own unpredictable advances -- what seemed like a good design for a reasonable pricepoint a year or so ago has been outstripped by advances in running the flowcells.

More Barcodes Please!

A serious limitation for some applications, particularly on the PromethION, is that the flowcell data generation capacity far outstrips what you want from one sample.  ONT has been lagging on providing deep sets of barcodes -- there's theoretically 24 for native barcoding but that's really a drop in a bucket.  Not only might you want to pack far more sample per run, but there's advantages to rotating your barcodes to deal with cross-contamination -- and the option of making libraries early and determining pooling strategies late.

Online Ordering Woes

A friend and former colleague of mine is at an exciting startup and on my recommendation are testing the waters with a MinION and are already thinking about GridIONs.  And are also echoing my complaints about the poor online ordering experience, between the split sites for ordering and tracking and the late arriving tracking numbers -- and unpredictable consumable delivery schedules.  Particularly with a huge marketing force soon to take over their rival PacBio, ONT needs to improve this experience and quickly.


A few LC's back -- I think it was two -- there was talk of the sensitivity of the flowcells, how only a tiny fraction of the library is ever used and many molecules never get near a pore.  If this could be fixed, then library preparation volumes could be potentially downscaled along with their expensive reagents -- as well as enabling libraries from vanishingly small amounts of input material.


ONT added a new feature to MinKNOW so that one can see the current rate pores are operating at, which is a proxy for the ATP concentration.  Groups are now using this to decide when to refuel the flowcell with ATP.  I don't know if this can be recorded to a file, but I'll repeat the wish that all and any status information be writable to an easy-to-parse file.  Just writing triples of label, timestamp and value would be enough.  I haven't looked if this is fixed, but my previous example was the duty cycle plots -- these are invaluable for troubleshooting, but the fact they were only available as an image in a PDF made it much harder to mine these.


The remarkable Cas9-guided capture protocol mentioned in San Francisco is out as a preprint, along with another scheme using Cas9 not only to make cuts flanking the target but to also protect the target region from exonuclease digestion.


Two throughput notes: I just saw the first tweet claiming 30 gigabases from a single MinION flowcell.  That's just bonkers!  Three such flowcells for a 30X human genome.  And the first 48-cell PromethION has now generated data.


The amazingly multitalented Ryan Wick set Porechop adrift a while back -- his further productivity couldn't be sandbagged by supporting every useful artifact he had produced.  That's completely appropriate and proper -- but Porechop fills an important gap in ONT's own informatics tools and they really should have a solid long-term strategy to support tool development and perhaps more importantly roll key bits like this into their own tools.

Hanging Myself Out To Dry

Well, that's it for now.  What did I forget?  What are you extra curious about in Nanopore land?

No comments: