Friday, May 15, 2015

London Calling Day 1: Highlights

Oxford Nanopore's London Calling conference kicked off today; I've Storified a large collection of Tweets from it, covering today up through about dinner.  I'll summarize some highlights below

Oxford had been promising that attendees would get their very own MinION; what hadn't been announced was that each one would have the attendee's name on it!

MAP Participant Talks

After an introduction from Oxford Nanopore CEO Gordon Sanghera, several talks by MAP participants kicked off the meeting.  While much of several of these talks has already been published, new details emerged as well. 

Notorious nanopore fanboy Nick Loman described his experience using MinION to analyze both a Salmonella outbreak and the ongoing Ebola outbreak in West Africa, including taking sequencing into the field.  In the case of Salmonella, in just 100 minutes of sequencing a firm conclusion was reached on the nature of the outbreak.

Yutaka Suzuki described using MinION also for disease surveillance, this time for Dengue fever, with an isothermal amplification (LAMP) generating material for sequencing.

Justin O'Grady covered the pressing need for rapid diagnostics in sepsis, in which identifying the type and nature of pathogen rapidly can mean the difference between treatment success and failure (an issue that once touched me personally).  Identifying from culture can require two precious days.  A key challenge for a sequencing-based approach is avoiding flooding the sequencer with background human DNA (an issue that arose back a few AGBTs in the DeRisi talk on encephalitis).  O'Grady pointed out that streaming analysis of sequencing data could really make a clinical difference here.

Sara Goodwin reviewed her group's experience with the MAP -- both highs and lows -- as well as their work on error correction and de novo yeast assembly.  

Several of these talks touched on issues with aligning nanopore reads to references or each other.  Goodwin uses BLAST, but LAST seems to still be a favorite for many others -- but nanopore-specific tools such as MarginAlign and GraphAlign are starting to emerge.

Jared Simpson covered his efforts to dig deep into the nanopore data for error correction and nanopore-only assemblies. This has enabled de novo assembly of E.coli into a single contig, as detailed in a preprint which has apparently just been accepted for publication.  He announced a new program will be available soon called eventalign which will directly align nanopore events to the reference, without using called bases.

Finally, Ewan Birney (who as always is very upfront about the fact he is a paid Oxford Nanopore consultant) described an effort called MinION Analysis and Reference Consortium (MARC) to assess inter-lab variability.  Birney was apparently giving out free samples of E.coli MG1655 to anyone who wished to join the effort.

Clive Brown Talk

The big fireworks were in the Clive Brown talk, which hit many of the points I had mentioned yesterday but also a number of things I failed to uncover.  A lot of upgrades to the ecosystem were promised to an enthusiastic crowd, ranging from such simple things as room temperature shipping of all components to much bigger things.

A key part of Oxford's system is an application-specific integrated circuit (ASIC), which is currently integrated into the disposable flowcell.  The ASIC listens to the channels and prepares the data for upload to the host computer of the USB cable.  The current ASIC has 512 channels; the next one will have 3000 and this will be the flowcell in the MkII MinION device.  With fast mode, the current ASICs will go from 40 bases/second to 500; the MkII devices are rated to handle 1000 bases per second!  Mk II flowcells are anticipated to be available in early 2016. Furthermore, the Mk II will feature a "Crumpet" design.  No, this doesn't mean clotted cream will now be in the running buffer -- instead the ASIC will be separate from a disposable piece which holds the pores and other fluid-exposed components.  This means that only relatively cheap components will be disposed, enabling even lower pricing than the current consumables. Crumpet devices are anticipated to have two week runtimes; standard practice now are run times no longer than 2 days (though Oxford has said they could well run longer).

Several of the earlier talks had remarked on sample preparation becoming a bottleneck in terms of time and effort  Brown mentioned a 10 minute library prep is in the works which will yield only 1D reads, which may be sufficient for some applications.  Another angle is to load the sample onto flowcell while bound to the beads used to purify the DNA; this is expected to enable sample inputs of 100 nanograms or perhaps even picograms.

However, the big splash in sample preparation is Voltrax (or is it VolTRAX? -- Clive's slides & Oxford's website use different typography), a device which sits atop a MinION and automates the entire process.  Voltrax will be disposable and programmable, with 6-12 sample input ports, and will be powered from the MinION.  Not clear from the tweet stream if any beta testing or launch windows were mentioned.

PromethION prototypes were apparently on display.  It is sounding more like the original GridION concept than the image I've had of 96 MinIONs ganged together.  PromethION is expected to mount-144,000 channels across 48 sensor chips.  Each chip will be able to run 4 separate samples. Multiple Voltrax units will be mountable on a single PromethION.

One aspect of Oxford Nanopore that is utterly different than any other sequencer in the marketplace is that each sequencing element can complete an input molecule and then start sequencing a new one -- this is a point that is sometimes forgotten.  So 512 pores on MinION MkI (if they were all active; 400+ active is considered a good run) don't just sequence 512 fragments, but depending on library preparation and quality and run time, can sequence many, many more.  For example, if you had a library of 400bp amplicons, then the current kits should be able to process an amplicon in 10 seconds per pore.  Now, there is downtime as the pore waits for another fragment to diffuse into its maw, but 400 pores can read a lot of small fragments or a smaller number of large fragments.  With the MkII flowcells, the number of pores is up 6X -- and PromethION will have 144K!  That will be a lot of sequencing firepower.

Brown mentioned a first application of an even more unusual feature of the Oxford platform -- by reversing the potential bias on an individual pore, the DNA it is processing reverses direction!  While this could be used to repeatedly analyze the same molecule, Brown instead described a different application using barcoded samples -- the flowcell could be programmed during a run to eject fragments bearing specific barcodes!  So, if you really want equal representation of all libraries, the instrument can solve the problem of poor normalizaiton by biasing the sequencing for specific libraries!

Fast mode should be rolling out in June, jacking the pore speed to 500 bases per second -- but with little or no decrease in basecalling quality.  Brown mentioned how the dwell distribution changes for the pores at the high speed in a manner which is actually advantageous for base calling.  I'd love to see him or Ewan Birney (who clearly gets this) explain this further; kinetics were a perennial weak spot in my biochemistry schooling.  The motor enzyme is apparently capable of running at 650 bases/second.  In any case, the basecaller is being rewritten for the fast mode.  Oxford claims to have achieved 12Gbases per run with fast mode, with 20gb targeted. On PromethION, 144K channels at 500 bases/second could yield 6.4Terabases/day!

What else?  Brown mentioned the issue of nanopores seeing every base modification or damaged base as slightly different -- a bug if you aren't interested in that (it depresses accuracy; I have some preliminary results showing that E.coli data has higher error rates in the vicinity of known methylation sites.  An interesting question there is how different backbones (e.g. phosphorthiolates - -either from oligos or in certain bacterial DNA) Brown also mentioned that Oxford has made progress on direct RNA sequencing, starting with DNA-RNA hybrids.  New pores (R9 chemistry) are progressing, with improved error rates.

What will this all cost you?  Oxford is planning a radical new pricing scheme, akin to pay-as-you-go mobile phone plans.  3 hours of MinION time is proposed to go for $270; with MinION MkI in fast mode that could be 2Gb, or plenty to sequence a number of bacterial genomes. Apparently a figure of $20 for the first hour was thrown out; that would work out to 5Gb of data with MinION MkII (all of these figures in this paragraph stolen from Nick Loman tweets).  PromethION will follow the same sort of model  -- no upfront cost to get access to the box!

Thanks to everyone who Tweeted!  I'll try to have a similar round-up of tomorrow's talks, though it probably won't be published until Monday due to a crowded personal schedule this weekend.


Rick said...

Sounds like there is no change to the requirement of doing the processing via the cloud; e.g., the need for fast internet on site. The idea of 'pay by the hour' sounds like the tie-in to the internet is even greater -- otherwise how would they know how much time you are spending?

Keith Robison said...

Rick: PromethION will have on-board base calling. I think the field epidemiology crew will be pushing hard to reduce the tie to the internet.

One model could be to just require license keys to go over the internet -- that wouldn't cut the tie entirely but would greatly reduce the required bandwidth.

But could also imagine some sort of license keys which you buy to enable a certain amount of sequencing on a specific device.

Torsten Seemann said...

Clive said the PromethION could be used with compute underneath, or without.