Monday, December 07, 2015

MinION Community Meeting 2015: Reflections & Wrap-Up

I spent the end of last week at the New York Genome Center for Oxford Nanopore's MinION Community Meeting 2015.  Since the family joined me for the weekend, I let my thoughts simmer for a wrap-up.  Plus I've been spending time scrutinizing the complementary "pen" for a USB connection and sample port, with no luck.  Wisely, I've given up on that -- so I can start the same process with the "notepad".  I've also finally stopped looking over my shoulder for a pitchfork-and-torch crowd after my numerous Twitter miscues, ranging from omitting speakers' names and affiliations to various mutations of the official hashtag (when I remembered any hashtag).  For a slightly better synthesis of the Twitter stream, see my Storify.
Yeah, the swag this time won't sequence. This fits with ONT's stated purpose for the meeting: so that MinION users can share enthusiasm, tips and successes; big platform announcements are reserved for the London Calling meeting each May. ONT did demo the VolTRAX microfluidics system which will provide a novice-friendly sample and library prep (initially only a 1D transposon-based library prep) by having it split, move and mix droplets of colored solution. A PromethION was available to see and touch, but beyond that the focus was on the current MinION environment, though with plenty of talk of some of still pending upgrades.

The fact that the transposon prep would generate only lower quality 1D reads (both in a standard format and on VolTRAX) was met by surprisingly little negativity; many users are running identification applications which work fine with only 1D reads or which currently use only information from one strand.  The What's In My Pot (WIMP) workflow on ONT's Metrichor platform appears to be quite popular, generating a constantly-updating fingerprint of what sequences have been seen in a metagenomic sample.

Perhaps one of more surprising talks was Dr. Zev Williams from Albert Einstein, whose practice focuses on couples with extreme fertility issues, particularly repeated miscarriages.  Many of these miscarriages are due to aneuploidies, so pre-implantation screening of embryos can be a route to a successful pregnancy.  A naive individual (such as myself) might not understand the value of speed in this application, but with current practice the embryos must be frozen while genetic results are awaited, whereas a sufficiently rapid protocol could obviate this step with its possible impact on embryo viability. The other surprising bit is that since relatively short sequences are sufficient to map chromosomes, and anything else beyond the amount needed to map superfluous, Williams has developed a very short fragment library method, which required adjusting adaptor-insert ratios and some of the incubations. 

Fast mode often came up.  Once slated for last summer, getting it out the door has proven a challenge -- Oxford told me that the higher data flow had stressed other parts of their system, revealing hidden issues.  So hardware and software had to be tweaked, but no complete show stoppers. Many speakers appear to be eagerly awaiting fast mode, doing proof-of-concept on applications that will require the higher throughput to truly succeed.  Matt Loose was the one speaker who half-jokingly said he is not looking forward to fast mode; his amazing MinoTour software processes data in real time, so clearly it will be stressed by substantially higher data flows. The other concern expressed about fast mode is that some groups are already seeing their solid state drives (SSDs) fill during a run, and SSD prices aren't falling fast enough to keep up with MinION development.

Loose demonstrated success with a nifty feature of MinION that seemed mythical: read-until.  In read until, software analyzes the read coming off the channel in real-time, and at some point can kill the read -- the voltage is reversed on the pore ("tasting" and "spitting out" as one audience member put it).  Loose showed how this can be used to normalize coverage of a set of amplicons and confirmed via Twitter that this can also be used to normalize by barcode, but numerous other applications await. Loose did underscore that a popular one is likely to be very challenging: filtering out human reads from metagenomic datasets.

Due to simultaneous sessions, I missed a number of talks that had good buzz on Twitter.  Several speakers in another session described exposing undergraduates (and even earlier?) to MinION sequencing.  Ebola sequencing in the field.  Rapid infection diagnosis.  Lots of good stuff!

A number of interesting talks on informatics.  One showed how MinHash can be pushed to enable species identification from metagenomes using Raspberry Pi-class hardware.  Another talk covered the GraphMap algorithm.  Another talk on methylation mentioned that methyl-C calling may soon be available in the NanoPolish program.  An area for active exploration, highlighted by multiple talks, is fast mapping of squiggles to references without first calling bases.

At least two groups reported library preps that are highly enriched for 20Kb+ reads; standard Oxford preps are in the 5-8Kb range.  A number of posters and speakers covered "ultra" reads, with multiple groups reporting alignable reads in the 150Kbp-200Kbp range, though not in large numbers.  ONT has started using the Sage BluePippin instrument to enrich libraries for long reads.  At least one group believes they have fragments approaching 400Kb, but can't get them off the beads.  

A big hit, particularly for me, was a live session in which a member of the ONT technical development team ran all the way through library prep, giving many tips and anecdotes about what ONT has and has not tried -- and plenty of feedback from the audience as well.  Much of this remains in developmen; ONT says they screen new ligases as quickly as they can get access to them. Most of it seems straightforward, though drying AMPure beads after an ethanol wash still seems like a very dark art (let them dry just enough!).

An interesting aside that came up in one of the informatics sessions.  The MARC paper highlighted a periodic (4 hour) switching of voltage in the instrument.  Given that this has a negative effect on sequence quality -- particularly for regions in the pores at the time of the switch -- its existence was enigmatic.  The explanation is that electrochemical drift in the flowcells occurs, and over 4 hours this adds up to about 5mV.  An early design decision in the hardware gave voltage control in 5mV increments, hence the drift cannot be corrected until about one step size of drift has accrued. Future hardware will incorporate 1mV steps, enabling more frequent but less drastic voltage adjustments.

Miscellaneous other tidbits: direct RNA seq on track for next year; multiple speakers pointed out the problems with cDNA preparation protocols highlighted by long-read sequencing - particularly the reluctance of reverse transcriptase to transcribe entire messages and the generation of chimaeras and similar artifacts via template switching.  Amplification free barcoding, and expansion of the long (24nt) barcode set to 96 distinct codes.  PCR amplicons don't require gross depth for accurate variant calling. Much more!

Between this meeting and the growing list of MinION publications, anyone who dismisses this platform as hype or a toy is delusional; MinION is here to stay. Applications are currently dominated by the special characteristics of the instrument: inexpensive (who would let undergrads touch their $200K sequencing box?), fast and portable. Read quality is improving with better models: the original HMM targeted 5-mers; now 6-mers and the current caller was shown here (and elsewhere) to make mistakes around methyl-C when trying to call to just 4 bases.  Conversely, pioneers continue to demonstrate a growing number of applications which can tolerate 1D reads and still yield useful results.  So Oxford is carving out a series of niches, many of which simple don't work well on any other existing sequencing platform.

However, should Oxford execute on the ambitious programme of improvements announced at London Calling, a radical new sequencing player would emerge -- and Oxford remains confident that these will all happen.  Direct RNA sequencing, as mentioned above, would greatly lower the cost-per-sample while greatly improving the quality of the results. VolTRAX promises to enable anyone, including all-thumbs greybeard bioinformaticians, to prepare samples for sequencing in the course of a coffee break, while not requiring the learning of arcane beadlore. Fast mode could mean sequencing even large bacterial genomes in a few hours.  PromethION is promised to be on track for release into the first field sites in early 2016, with potentially frightening throughputs (terabases per day!). 


Dale Yuzuki said...

Thank you for the informative post - the VolTRAX looks remarkable, the idea of reversing the pores also sounds intriguing, and it's great to see how much progress has been made in the 2 years since some of the first data was presented by the Broad at AGBT.

On the pre-implantation genetic screening for aneuploidy, did the speaker address the number of reads needed per sample for detection? If I remember correctly on the PGM (a few publications by Nathan Treff in New Jersey, Dagan Wells from Oxford come to mind) they used on the order of 30K - 50K reads. And FYI Ion Torrent launched a kit around this application, called ReproSeq, for the assisted reproductive technology (ART) market.

Thanks again for writing this up - exciting time for them, and while I was at AMP (Ass'n for Molecular Pathology) and noticed them there it's clear they are going after the infectious disease market. Will need to look at their recent posters / publications to get a better handle on that.


Jonathan Jacobs said...

"One showed how MinHash can be pushed to enable species identification from metagenomes using Raspberry Pi-class hardware." -- ARGH! There must have been something in the ether in the days leading up to MCM. I wish I had been there...

Keith Robison said...

Dale: According to a tweet of mine, Williams said 15K reads was sufficient to call aneuploidies -- in practice he got 2X the number of reads he needed to call Chr21

A longer reaction to your excellent question as a full post sometime soon (about 1/2 drafted)

JB said...

Being quite new to the field (and your blog!), I might address a topic that is not in the spotlight atm, but, was there any talk about the development of solid-state-nanopores? I know ONT have been looking into e.g. graphene as a material, and other groups Molybdenum disulfide, making a more cost-effective and robust system by replacing the chemically more sensitive pore-making proteins.

Keith Robison said...

JB: I confess to not following the research on nanopores themselves very closely -- there has certainly been a lot of work on this, but it's hard for me to sort out what is anywhere near commercialization