28 min read

London Calling 2018

Gordon Sanghera - Opening

Intro music - Arctic Monkeys

  • A band that was around when most ONT staff were alive
  • In 2005, they went straight in at number 1
  • No sales, no marketing, used myspace to allow people to download one sing
  • It was a bunch of kids
  • Since then, there's been no need to pander to the large organisations

Sequencing landscape

  • So many things have been proposed for nanopore sequencing
  • The oil industry has problems with microbes
  • The shipping industry has problems with ballast tanks

Statistics

  • The conference has 600 people - 500 customers, and 100 Nanopore people (with blue lanyards)
  • Goal of ONT: Allowing anyone to sequence anything anywhere
  • There are 7,000 MinIONs; 140 GridIONs; 40 beta PromethIONs; all done with community-based campaigns

What it all means

  • We are at the cusp of the fourth industrial revolution
  • We are seeing things that are in sci-fi films
  • Uber's actual goal: delivering anything to anyone anywhere
  • We are bringing testing closer to the customers
  • People can use Air B&B instead of hotels
  • With nanopore, we are enabling the Internet of Living things
  • There are 12500 sequencing centres in the world
  • When people are provided access to real-time tools, we open up new opportunities
  • Genoscope has 85 coral species
  • General theme: More people, more diverse applications

Service Provision

  • KeyGene is the first authorised PromethION service provider
    • Providing service to the lettuce industry

Zoe McDougall

  • The Dunbar Number - 150; about the number that were in the first nanopore conference
  • Need to work out how to help people
  • Ninja programming schedule
    • Repeating some breakout sessions
    • A workshop to improve the digital community (Andy Davies)
    • 300 people have downloaded the software App
    • Interact button allows people to ask questions during plenary sessions

Angela Brooks

Video

  • Every cell has the same genome, but different cell types have different functions
  • The transcriptome is what distinguishes cells

Polyadenylated RNAs

  • Have non-coding as well as protein-coding
  • Can be edited or modified; over 100 types of modifications exist
  • Limitations of short-read sequencing
    • Fragmentation
    • Conversion into cDNA
    • doing PCR (even a problem with long-read cDNA)

Nanopore Sequencing

  • Sequencing is 3´ to 5´ (not typical)
  • RNA consortium was formed to look at human cell lines
  • 6 universities sequencing GM12878
  • The same cell culture was sent to five institutes, a separate culture was done by Birmingham
  • Can go to github to see the data
  • 13 million native RNA reads, 24 million cDNA reads

Observations

  • Native RNA was longer than cDNA
  • Error rates about the same
  • Good correlation in gene expression levels
  • Also good correlation with Illumina
  • Birmingham clustered separately when using PCA
  • Same sample clustered well together

Read fragmentation

  • A lot of reads that seemed to be fragmented
  • MTCO1
  • Read length doesn't drop over time (so there's no fragmentation happening within the flow cell)
  • 45% of the reads were full length
  • 20% were partial due to artefacts (e.g. basecaller truncating the read)
  • 35% other partial reads, maybe due to sample prep, or degredation of RNA

Full-length assembly

  • Using orthogonal information to distingush full-length transcripts
  • e.g. ChipSeq - looking for the promoter region. If not correlated, then it's probably not full-length
  • Using short-read support for read correction
  • Can look at full length and compare to annotation
  • Can pick up novel transcripts
  • Largest high-confidence transcript: 10313 nucleotides with 48 exons

How many sequenced reads are needed?

  • At about 5M reads, the number of transcripts start to plateau
  • Transcripts keep getting detected at high coverage (more than 75k transcripts)

Challenges

  • Allele bias
  • Also looking at polyA tail lengths [see polya_estimator]
  • Generally 30-150 nucleotides
  • Detecting m6A, found that it was kmer-dependent
    • More differences for some, but not others
  • Remaining challenge: hard to distingush full from partial RNA

In-depth analysis of 100 reads

  • Done manually by Mark Akeson

Questions

  • Looked at squiggle for capturing the 5´ cap (methods in progress)
  • Using minimap
  • Longest read from titin (aiming for 100 kb)

Dan Turner

Video

ONT Applications Team

  • Team in Oxford, New York, San Francisco
  • Oxford: sample prep
  • New York / San Francisco: showcase
  • Concerned with what's going in and what's coming out
  • Not all library prep options are good ones
  • Need to be clued up about the analysis at the start
  • Vision: End-to-end workflow selector

Worked example: Structural variation across the rabbit genome

  • First question: what is your end goal / aim?
  • Priorities can be re-ordered
  • Options with crucial information
  • Presented with extraction options (e.g. Qiagen genomic tip)
  • Then bioinformatics
  • Then a PDF protocol is created
    • Bespoke, not about how to run the MinION
    • All extraction protocols available on the website

Sample prep concerns

  • If analysis can't use short fragments at all, beter off getting rid of them
  • Working on syringe-based method
  • How clean the prep is is important
  • There's now a table of common contaminant concentrations
    • For example, can get up to 20% ethanol contamination (i.e. could get away with wine)

London Calling Conference Demonstrations

  • ONT is sequencing Sarasin's minnow
  • Want to generate about 100Gb reads, read N50 of about 15-20kb
  • Including control species
  • Doing library sample prep on the VolTRAX

VolTRAX

  • Portable sampling and library preparation device
  • New video: devices in a backpack
    • Find some poo on the side of a track, curious to know what it is
    • Excess liquid into binding and lysis buffer
    • Feed through ecotip to get rid of cell debris
      • Was too clean
      • Dry out
      • Elute from filter
    • VolTRAX; load DNA and other reagents
    • Remove sample from extraction port, ready for MiNION
    • Use FASTQ file for BLAST
      • Discover that the poo sample was from a rat!

Direct RNA Sequencing

  • A quick plug for the paper; see Dan Garalde

Tombo Software tool

  • Signal deviation gives evidence of modifications
    • "I don't know what the modification is, but I can see that it is modified"
  • Now supports m5C
  • Can also see DNA modifications
  • AUC for m5C: 0.99; m6A: 0.98 (compared to existing method)
  • Still optimising Tombo to work with the PromethION
    • Have found CHG methylation, as well as CHH methylation

Structural Variation

  • There are more structural variants in the genome than SNPs
  • Duplication detection works with nanopore sequencing
    • e.g. 1kb stretch, finding six copies in tandom
  • Does it detect structural variation that isn't there?
    • Some sequence contexts don't work with 1k chunks for base calling
    • With 10k chunks, there is 90% support for found variants
    • Ideally, want to do sequencing without any chunking at all

Pore-C

  • Long-read chromatin confirmation capture
  • Using long reads, not short reads
  • Get a lot more information about interactions
  • Plot the interaction map
  • Only did with 1.5X coverage
  • Can use information to verify assemblies
  • The last of the analyses have been done

Breakout -- Assembly

Antony Bolger

Video

Todd Michael

Video

Danny Miller

Video

Panel Discussion

  • Cheaper / easier to use nanopore + Illumina; only need to use 30X for nanopore
  • Nanopore + Illumina + Optical mapping
  • Nanopore + Pore-C
  • Should we error-correct reads prior to assembly?
  • Canu still does metter on mammalian-type genomes, probably due to quality
  • Reads passing the filter (Albacore/pass) produced the best assemblies
  • If filtering, removing shorter reads is better than best quality reads

Lightning Talks / Day 1 (James Brayer)

Logan Mulroney

[No video?]

  • Direct RNA sequencing issue: can't catch the last few bases
  • When strand is sequencing through and motor lets go, bases move through quickly
  • Add a few bases onto the end of the read
  • Done on yeast RNA
  • Helped by Dan Garalde, Mark Akeson

Kai Sohn [poster #1]

Video

  • Superman, Pope John Paul II, Muhammed Ali all died from sepsis
  • Most methods to identify pathogens are failing
  • Using cell-free DNA to identify pathogens in the blood
  • Clinical cohort, 11% positive, Illumina 71% positive
  • Time to result: Illumina sequencing took 16h
  • MinION is not designed for cell-free DNA, DNA is mostly of human origin
  • Working with 5ng, more increased throughput about four times
  • About 50% of the reads were accurate enough for a diagnostic measure
  • Find reads from infecting pathogen within minutes; sequencing within 5h with clinical samples
  • Better to collect the sample, but no clear sepsis marker

Jennifer Idol [poster #35]

Video

  • Works in a sequencing core, not a research lab
  • Three samples from 1000 genomes project, needs 6G to [end out] the Puerto-Rican
  • Uses optimised version of 1D kit
  • Phenol-chloroform with Qiagen MaxTract tubes for long reads
  • Have 30X coverage for all samples
    • 460/600μg DNA GM19240, HG00514
    • N50 40kb, 30kb
    • Normal yield for about 4Gb per flow cell
  • got 30X in about 2 weeks on a GridION
  • Fresh extractions a lot better than old extractions
  • Have new protocol with max read length up to 450kb
  • Cost of 1 human genome at 30X was about $20-30,000

Nick Schurch [poster #29]

Video

  • Anya and Kasia are amazing Direct RNA Sequencing wizards
  • Want to look at reliability, replicability
  • Very similar to humans in terms of transcriptional complexity
  • Consistently getting 1-2 million reads, 1.4 million aligned
  • RuBisCo Activae, extremely highly expressed, 52,501 alignments
  • Good splicing structure, most splices are novel and not supported in the official annotation
  • To get to 2 million reads, use 1μg RNA, be super-careful in lab prep
  • Consistently getting 1-1.2 million reads, about 850k aligned reads

Wouter de Coster [poster #17]

Video

  • Using PromethION in February
  • Currently highest yield on PromethION (98Gb)
  • NA19240 (Yoruban reference genome)
  • Well-characterised
  • 5 Flow cells, not all fantastic [59, 50, 29, 29, 72]
  • Shearing gives substantially higher read yield
  • Shearing gives a tighter read length distribution, shorter tail
  • Making data public as PRJEB26791
  • Blog post available now
  • Developing workflow [nano-snakemaker]
  • Sniffles detects more structural variation than nanoSV

Olga Francino [poster #11]

Video

  • Looking at microbial samples
  • Using 16s variable regions
  • Can we reach lower taxonomic levels with 16s?
  • Full-length 16s (1500bp)
  • Tested with mock community
  • Tested with mastitis buffalo / milk samples
    • 5 healthy, 6 subclinical
    • Completely different betweeen the two samples
    • 75% of the sequences were categorised down to species level
    • Infective agent was Staphylococcus aureus
  • [Excellent workflow image]
  • 12 samples, need a flongle now

Clive Brown

Video

Novel statements selected by David Eccles

  • Live sample prep, sequencing, and WIMP analysis of Clive Brown's DNA by Ruth Moysey
  • Voucher for Series D ASIC; longer run time, potentially 30 Gb from a single MinION
  • ONT has a roadmap for moving to 500 Gb from a single PromethION flow cell (NovaSeq-busting prices)
  • [Read-until is coming back [in anger]
  • MinIT is orderable in the store
  • Mk1C sequencer (MinION combined with MinIT) should be available December
  • Epi2me will be embedded in MinIT, GridION, PromethION (local analysis workflows)
  • Epi2me will be commercialised via MetriCoin (workflow currency, included with flow cell / kit purchases)
  • New 1D² kit coming that uses UMIs in adaptors, increasing accuracy to Q20 and allowing amplicon sequencing
  • Miyagi tool for combining reads from different pores for increased consensus accuracy
  • Slightly mutated forms of R9.4 have been tested; could release a variant-pore flow cell in a few weeks if there is interest
  • Flongle early-access begins now, released commercially in Q3
  • Zumbador's official name is Ubiquibopsy (or Ubik tube)
  • V2 VolTRAX (heat, PCR, 10 samples) is available for ordering in the store
  • New SkunkWorx technology: a combined MinION + VolTRAX, dynamically embeds pores between droplets
    • Too much novel stuff to summarise, see here

Introduction

  • Next year O2 arena -- looks like a nanopore, fits 20,000 people
  • Every year, I have been told that I need to do this: promised investors I would do a live demo.
  • Have been spitting in tubes, but soomeone else will do the prep.
  • Ruth Moysey takes a tubefull of spit from Clive.

Recap - get into rhythm

  • Aim: Anybody to sequence anything anywhare
  • Trying to make runnable outside labs
  • real-time full-length reads
  • ultra-low cost.
  • In the past day, a few people have been demonstrating all features.
  • This was a dream a few years ago.

Olbigatory nanopore slide

  • An array of membranes on a chip that we designed
  • catch DNA from solution
  • stream through the pore
  • generate tiny signals (a lot of work decoding) All asynchronous
  • get a full-lenth read in a few seconds
  • Electronics can be made small, cheap
  • Design breaks the traditional mould - Would be stupid to make an electronic device with fluidics
  • Also key for ONT was designing array in the factory

Other talks

  • There have been a number of previous talks [by Clive]
  • recently spoke about sub-1000 sequencing run; mostly on PromethION ($600)
  • GridION conceived Feb last year, has been relatively successful

Platform

  • Product line as it stands, all in-field
  • MinION is the most established
  • Out there GridION is doing well; easy to own, people using it as service
  • PromethION - a beast of a machine, now out, runnable, generating pretty high yields
  • Over 1000 Customers for GridION; We think we can ship within a week of an order.
  • Certification program, growing quite rapidly

Grab bag

  • In your bags is a token that gets you access to a slightly modified version of the MinION flow cell: a series D ASIC
  • It's just better
  • First design had a few faults, including current cross-talk; all things ate into yield
  • This is a fixed-up version of the ASIC
  • Can now enable very-long run times with sustainable yield of up to 100 hours
  • Enables 30Gb on a MinION

PromethION

  • 48 individually-controllable flow cells

  • 144,000 concurrent independent nanopore channels

  • Can hot-swap

  • Designed for quite lumpy workflows

  • Can be run as a high-throughput sequencer, just like any other.

  • Now dependent on Ruth on the Maracas

  • Now dependent on You lot for reporting on what's happening on PromethION

  • Now out of the Early access phase; $160,000 starter pack

  • Can take box back

  • When you buy PromethION can have access to upgrades

  • Next version will run 48 flow cells

  • Have shipped 40 PromethION by this conference

  • should get to 60 by end of June.

PromethION Yields

  • What do new customer yields look like?
  • Most getting over 50 Gb (this was the target)
  • Some getting over 60
  • ONT is getting internally over 170 Gb
  • A roadmap to move to 500 Gb per flow cell
  • These are NovaSeq-busting prices, provided you get the yields.
  • New flow cell: single channel. Also a redesign of the multi-channel flow cell.
  • Will make sure that all of MinION kits will work on PromethION (RNA coming in June)

MinKNOW

  • MinKNOW upgraded
  • Very nice UI, innards of the GUI have been completely refactored.
  • Thinking about things like read-until
  • Including progressive unblock & dynamic voltage control, should increase data quality.

Progressive unblock

  • Blockages when you shove DNA thought a hole. Single strand forms structures on the other side of the pore.
  • Can unblock by inverting potential.
  • The old software was the simplest thing that worked
  • Unblocking was quite violent, would remove channels
  • The new version doesn't destroy the channel; a more subtle, sophisticated version of unblocking the pore
  • Probably a larger uplift in difficult samples.
  • Have generally improved kit quality

Software

  • 3-4 years ago pulled in software team to talk about Read Until

  • As DNA is going through the pore, can basecall and do something with it before it's gone through the pore

  • An in-silico selection

  • We wrote an API.

  • Entire innards of MinKNOW have been rejigged, read-until is coming back

  • balancing amplicons

  • Sequencing a certain region of human

  • Coming back in second-half of year, coming back in anger

  • Will be able to dynamically select molecules on the fly

MinIT

  • For MinION: Common complaint is the computer; that has caused a lot of trouble

  • Making a device called MinIT

  • It should be almost zero configuration.

  • Plug it in, connect it, pair your iPhone display, ready to sequence.

  • Can run off batteries. Can run off 12V supply.

  • Will put Guppy in there, will keep up with MinION basecalling, critical for ReadUntil

  • Also embedding Epi2Me.

  • Pretty confident that a large number of people will want a hand with analysis

  • MinIT Should be orderable today.

  • Have gone for GPUs in MinIT; going for neural-network based AI

  • A bunch of sequential matrix computations

  • Quite a large number of bioinformatics things can be optimised on the same hardware with the same performance.

  • MinION Mk1C - MinIT with the aility to load a flow cell, or flongle

  • Target December

  • Should be able to run a single flow cell without any external needs

  • It's the kind of thing a dentist can run

  • It also looks really cool

Epi2Me

  • A bioinformatics platform

  • Embedded workflows doing the kind of analyses that people want to run

  • Can be run in the cloud, or workflows can come to the computer: GridION, PromethION, MinIT.

  • Lots of reasons, 1) a lot of computing in the boxes.

  • Expertise in optimising on GPUs

  • Sometimes people don't want to pipe data over to the over side of the world

  • We think it's more usable now, can upload custom reference, will provide data storage

  • We will release Epi2Me on MinIT in Q3 - a MinIT with MinION will do this locally in Q3

Epi2Me Commercialisation

  • Will be commercialised
  • When you buy a flow cell; have invented something called MetriCoin
  • When you buy flow cells or reagents, will get MetriCoin, can use to run workflows
  • People who don't have bioinformatics support will have access to bioinformatics services

Chemistry improvements

  • Platform performance is now much closer to what Clive tweets
  • Customer runs by kits is converging, variance is decreasing
  • Largely due to kit simplification
  • Early kits had a cholesterol tether; it sticks to everything that is hydrophobic
  • We have been removing that from all of our kits.
  • Generally customer performance has started to converge with nanopore.

Read Length

  • Fragment length is read length; largely in customer hands

  • Most platforms, the optics / chemistry conks out

  • If you can present with long reads, it will sequence it

  • Some of thought-leading developers have tried to present pore with enormous, intact lengths of DNA

  • The developers found a bug in the software, ONT software was chopping up reads

  • In principle, if you can do 2 Megabases, can do 20 Megabases

  • Challenge is whole-chromosomal sequencing

Yield Drop-off

  • Initial software that classified data was the simplest that worked

  • Software misclassified reads, which would degrade the data

  • Low-complexity regions, rectified in next software release in June

  • Need to Scale data, scaling can be biased, to be fixed (fix went out 2 weeks ago)

Unblocking

  • Extracting from chitinous insect won't be the same as a drop of blood

  • Not the whole story, can get DNA out

  • There have been issues with things sticking to tubes

  • If initial yield was high, would drop off suddenly

  • Chickens do this

  • Software that you need to kick out DNA is different for different samples

  • Some kinds of DNA form secondary structure as they get to the other side of the pore, creating a block that kills the pore.

  • Another upgrade coming to change the unblock logic that should level out yield differences over time.

Kits

  • Recently settled lawsuit about hairpin

  • Came up with a system to do 1D²

  • Version 1 went out a few months ago

  • It is now due for a revamp

  • Adapters will be improved to imporoved the performance of the system

  • Looking for 60% reads 1D², should get to Q20

  • Software is antediluvian, 4 years old

  • A lot of people have complained that they can't do amplicons

  • New kit optimised for PCR uses a special set of barcodes that will enable barcodes to be paired up (UMI).

  • Will go back again, revamp it, and make a performant version.

Targeted sequencing

  • Technology was to take a deactivated Cas9
  • Guide RNA, hybridised to sample, subset of reads is pulled down to pore
  • Working quite nicely
  • Will be out as a kit later in the year.

Direct RNA sequencing

  • Can put RNA through the pore just like DNA

  • Proven very popular, there's lots of stuff in RNA

  • Not as performant as DNA product

  • Get full-length RNA molecules, can see all the modifications

  • Can really get to the full-length biology.

  • Nanopore is quantitative, see all the splice formas.

  • Now focusing on making RNA as performant as DNA

  • Want to increase speed RNA goes through the pore

  • Will move to 110 bases per second

  • Looking at improved adapters, simpler, fast library prep, incremental advances in basecalling

Field sequencing

  • Despair when looking at what people take to the field - polystyrene ice bucket kills me

  • A lot of work has gone to ambient shipping

  • Looking to post flow cells without cooling

  • Can ship at ambient, most reagents

  • Looking at storing ambient, a few kits left to lyophilise

  • Especially an issue in places where don't have a lab

Consensus Accuracy

  • Looked at fixing this issue a few months ago

  • The thing that most people hold up to nanopore

  • Cornering the rat, and then stamping on it

  • We're now tickling the rat

  • We're not at the limit yet

  • Worth talking to the team who are here

  • There are some limitations in the current software

  • Current training set, issues with scaling / chunking of data

  • Still aren't learning all the context

  • We still train on bacteria

  • We don't learn in damage and modifications

  • There are some software limitations, and other issues.

  • People assemble and polish, not fully-utilising all the signal

  • People don't exploit alll the information in the signal

  • Quite a lot of work to do.

  • Bases are not modelled, similar current for different sequences.

  • Progress on homopolymers, over past few months.

  • Nanopolish is still the gold standard for treating these data, but medaka can do well in some contexts.

Nanopore solutions

  • Have medaka, only works in base-space

  • Comparable to nanopolish, but much faster

  • Have been working on a new tool, miyagi, for correcting homopolymers

  • HMM based, that's the way to do it

  • comparable or better to nanopolish in some circumstances

Fixing the problem

  • One way: improve the chemistry.

  • People are alreads combining data with another certain company

  • Has been long on our agenda to do the same thing on one platform

  • Would then not need the other company

  • If you could combine multiple pores, would get flattening out of errors

  • R9.4 pore signal comes mostly from 3-4 bases

  • A run of Ts bigger than 3-4 gives flat signal

  • Can estimate bases using time domain

  • Would be better if we could de-flat the sections of data

  • Using read-ahead to look at more bases, rather than less

  • Will always be averaging over more bases

  • Deal with more bases using machine learning

Pore changes

  • Use Pore R8 - Lysenin based

  • R10 has two points about 9 bases apart

  • A particularly pathocogical error is a round of 5 Ts followed by two Gs

  • R10 spans the homopolymer giving an uppy-downy signal that the software can decode

  • Equally, number of percentage correct homopolymers with R10 is better

  • Think that green accuracy line will shift up.

  • Can generate orthogonal errors that can shift consensus

  • These are now under very active development as potential products

  • What about combining together?

  • 50-fold R9 with 50-fold R10

  • can easily exceed Q40 in consensus, proof of concept

  • These methods are being implemented in miyagi tool, can use yourself (if you had R10 data).

  • Still not using dwell in these data

  • Using time domain, can boost these further.

  • The principle is shown.

  • Why not 5 pores?

  • Why not 500, each with completely different error modes?

  • We're going to crack this this year, and get it out to you as soon as you can.

  • Doing very well on variant calling, specifically on SNVs.

  • Have made at least one mutant of R9.4; R9.6. 9.4 gets one homopolymer wrong, 9.6 doesn't.

  • Can make even variants of 9.4 that have significantly uncorrelated signal. Could release this in 2-3 weeks.

Template changes

  • Could mess with the complement of the sequence

  • About 20 minutes, a pot of nucleotides in the kits, swap T for U and train basecaller

  • A/C/G/U version has different errors from A/C/G/T version. More complicated.

  • My preference is to put multiple pores in the chip.

  • Can provide these tools to allow people to mess with the pores.

  • Combining all three: two pores plus complement

  • Combining data that isn't correlated, as long as you pick the right data.

  • Will keep 9.4, but could make 9.6, could make R10

  • Question: is this useful? is that of interest? When would you like it? [let ONT staff know]

Flongle

  • It's coming, really soon

  • MiNION + 1-off adapter, that's the flongle

  • Has most of the electronics that's in the MinION

  • An array of copper electrodes

  • Can then make a very cheap crappy plastic flow cell with magical chemistry

  • Separates the cheap bits from the expensive bits

  • 128 channels, limited by the forces that are needed to make connections

  • The data is the same, can flongleise a MinION

  • Can flongleise a GridION

  • Projecting that v1 up to a Gb from one flow cell, should get 3Gb from a flongle.

  • People don't want lots of data, problem is about how quickly enough data can be obtained.

  • Also good for bacteria and virused

  • People should buy more if it's cheaper

  • Flongle works, just like MinION

  • Squiggles are squiggles, they are the same

  • Invites for early adopers have going out

  • Looking at $90-$100 per flow cells, about 5-10% of the total treatment cost for dentist

  • Early access begins now, will release commercially in Q3

Where to go

  • Drive is to get sequencing out of the lab
  • Can survey what's going on in a river
  • Cheaper, easier, package it up, remove any extraneous equipment.

Zumbador

  • Clive had the largest office, but had to rename Zumbador.

  • Idea is that you can take something very cheap, a piece of plastic, introduce a liquid because they're easy

  • After some reasonable time, like 10 minutes, can just sequence what's in there.

New name: Ubikwibopsy (Ubik tube)

  • Ubiquitious

  • Kwick

  • Liquid

  • Biopsy

  • Turns out that DNA is really informative

  • Don't want to ship samples off to california

  • Want anyone to be able to do this themselves, and have complete control over their own data.

  • Video: can put spit in a tube, a bunch of stuff happens in the tube, can bind DNA onto solid phase support, then separate. It drops onto the flow cell.

  • Have embedded the library prep onto the flow cell, comes ready to run.

  • DNA in, sample prep appears early in the pore

  • Can get meaningful data really quickly

  • Yields aren't massive, but we've got it

  • 10 minutes from spit prep to WIMP

  • Had bacterial infection, could fix it, see a drop in the bacteria.

  • Saliva, lysis 10 minutes on Ubik tube.

  • Dentist offered to buy one.

  • If you encapsulated enrichment, could use read-until to exclude human, or include human.

  • We Might not be the right audience for that.

  • Future versions like flongle and smidgION will move towards anybody being able to touch a sample to a flow cell and know what's in.

  • Metrichor will make quantitative analysis of the self available direct ot the consumers.

VolTRAX

  • Version 1 a while back.

  • Main use is the proper version, version 2 ready to go out a bit later this year.

  • Supports all kinds of complex lab workflows.

  • 15 inlets.

  • Can do heat, PCR, fluorescence detection. It is a lab on a chip.

  • Starter pack for V2 $8,000

  • As I speak, the shop is open for orders, can buy a VolTRAX v2.

  • Will prioritise people who got a V1.

  • Can do up to 10 samples, 2xGridION.

  • Pricing starts to get quite attractive for multiple samples.

SkunkWorx

  • A few years ago, pulled people into my office, asked them to do a thing.

  • Amazingly, people went off and did it.

  • A VolTRAX - MinION hybrid, amalgamating into a single device.

  • Now have another DNA sequencer that can be made.

  • Completely automated, very complicated droplet-based workflows can be built in.

  • No physical handling, can work directly from cells.

  • Form dynamic membranes between droplets on the VolTRAX.

  • Can insert a pore, and have a sequencer.

  • Move the droplet to a certain part of the array and droplets look like a MinION channels.

  • Squiggles are the squggles; they are the same.

  • Droplets can be quite small, or quite big.

  • Can swap the droplet.

  • Can access the trans side.

  • Can take the things you sequenced, and PCR them, or sequence again.

  • Target is to get to 128 channels.

  • More likely to be less than that.

  • Currently 8-channel breadboard.

  • Extraction zone, Library zone, sequencing zone.

  • Henrietta Lacks cells being aligned and lysed.

  • No shearing; what if you could just get whole chromosomes and put it between two droplets.

  • Lots of things you can do.

  • We tend to stick DNA to the membrane.

  • Can coat up the droplet with DNA, DNA whizzes around and will find pore within seconds.

  • If you have a tiny amount of material that you want to mine digitally, can mine out that cell.

  • With read-until, can eject high-abundant stuff.

  • Schemes where you can re-pair sequenced DNA and put it through again.

  • Can also just hybridise things to DNA.

  • Some Cas9 are SNP sensitive, can count thousands of presence - absense.

  • Spit blood in, all in one, very flexible measuring device.

  • We have the proof of concept, all now about making the box.

  • Will be coming your way; not talking years.

  • The mind boggles of what you can do.

  • A small platform at relatively low plex. Our next likely major product launch.

Timeline

  • Revision D ASIC, can get now. Probably broadly by middle of Q3. Should see 20-30 Gb.
  • Kit9 much improved ligation kit
  • PromethION went on commerical release today
  • PromethION flow cells orderable
  • Epi2Me commericalisation
  • Software umblocking fixes
  • MinIT shipping in July
  • Ambient shipping in July
  • First flongle shipments in July
  • Voltrax V2 in August
  • Much more performant 1D² in August
  • RNA upgrade in terms of yield / quality in September
  • In December fully-performant PromethION (48 flow cells)
    • ONT has just got 3Tb internally
    • target is 12 Tb per run, equivalent to 3-4 NovaSeqs
  • December Mk1C MinION

Questions

  • Flongle flow cells should be able to be chuckable; throwing away electronics is difficult.

  • Multiple pores - Coverage will be reduced because errors aren't correlated. As long as you can pick the right one, which is the high and low quality base.

  • Trying to get the same kits and settings on all devices.

  • SmidgION - We can make it, that's not a problem. ASIC is reallyy cheap. Flongle comes first.

  • R10 pore - frequency of homopolymers. Current software can call homopolymers as -1, some bias in software that is consistently doing it wrong. Will get a lot further with just incremental improvements.

Lightning Talks / Day 2 (Thomas Bray)

Franz Josef Müller

  • Nanopore sequencing of short repetitive DNA
  • A number of short repeat disorders (e.g. fragile X: GGC repeat, usually 30-50, but can expand up to 200 repeats)
  • Why nanopore sequencing? Looking at kilobase stretches of 100% GC sequence, southern blot remains a diagnostic tool
  • Good news: raw signal allows tracing / counting cases
  • Regardless of the system, no real ground truth of the repeat number (in the normal case)
  • Used nanopore-simulation (crohrandt)
  • Pulls data, generates raw signal
  • Entered defined repeat expansions based on the ground truth
  • Made another tool: nanostrike
  • Uses a signal model to identify repeat boundaries, quite good agreement with manual counts
  • Cpf1 enrichment
  • Poster actually describes a different tool

Natacha Couto

Video

  • University of gronigen
  • Looking at vancomycin-resistant enterococci
  • Intrinsic resistance to several microbial classes
  • VRE dissemination can exchange mobile genetic elements
  • Colleagues studying 36 isolates
  • Did Illumina sequencing, found 7 clusters
  • Some strains were clustered differently, but isolated from the same patient, or the same ward
  • Wanted to look at transposons
  • [Showed photo of PI - John, because he kept showing photos of her] -- she did the sequencing herself, it wasn't John's work
  • No shearing, assemblies using Unicycler
  • Could see transposons, look for differences, two on the chromosome, other two in a plasmid
  • Looking back at the tree, found that core genes could be linked by common transposons
  • Important to study structures, combine short-read sequencing with long-read sequencing

Stefanos Siozios

Video

  • Works with microbes and arthropods
  • Microbes that pass from mother to offspring
  • Image of maternal transmission to oocytes, paternal transmission as well (but rare)
  • Biologically important, can alter vector competence
  • Can, for example, kill of male offspring
  • Has been hard to complete genomes
  • Only a minority of microbes can be cultured outside the host
  • Microbes have highly repetitive genomes
  • Repetitive content is difficult to resolve
  • Trying to make an improved reference genome (estimated size of about 3.5 Mb)
  • Large amount of extra-chromosomal elements; PacBio also failed
  • Two different library protocols. Very long reads provided the ability to assemble the main chromosome into a single circular contig - got extra-chromosomal elements as well

Rainer Waldmann

Video

  • Likes idea of VolTRAX-based single-cell sequencer
  • Doing Illumina sequencing, need to fragment DNA, end up with only 3´ sequences
  • Problem with amplification, need to correctly identify UMIs and cell barcodes
  • Clustering gets easier if you know the barcode + UMI sequences
  • Did 1 nanopore run with 190 cells, 30M reads needed for 5X UMI
  • Did 951 cells using targeted nanopore sequencing
  • With one cell: reads with the same UMI should be the same molecule; some outliers
  • Nanopore data can be linked to Illumina
  • Good idea to use Illumina data with single-cell data

Yutaka Suzuki

Video

  • Sequencing core laboratory
  • Started analysis with cancer cells
  • 9 patients, some initial success for cancer results
  • Used the same process to genotype malaria patients
  • Making a geographic distribution of modifications
  • Moving to Papua, needed to modify protocols
  • Replaced PCR with isothermal amplification
  • Expanded to global G-RAID meeting
  • BBC (UK) found out about this
  • Wanted to do the first human cancer genome
  • With the field kit, first do Malaria in Indonesia

Markus Haak

Video

  • In spare time, likes to detect unnatural bases (e.g. isoG, isoC)
  • Detection method should be PCR-free and polymerase-independent
  • MAX: looking at different restriction enzymes, method applicable for certain sequencing contexts
  • Developed iCG ased on sequencing reference DNA samples
  • Unnatural bases have substantial influence
  • Automated model, does event correction
  • During event correction, bad events are redistributed
  • Data points are based on Dynamic Time Warping, shift to correct misalignment due to unnatural base calling
  • Quite potentially adoptible to other bases or base calling
  • Haven't tested against existing modified natural bases

Liana Kafetzopoulou

Video

Intro

  • ONT managed to fit her name on a MinION two years ago
  • Want to be able to sequence all RNA viruses with a single protocol
  • Charles Chiu published a protocol: reverse transcription using a random nonamer [9-mer]
    • creates a cDNA library

Samples for testing

  • RIPL
  • Samples that were the most prevalent: chikungunya & Dengue
  • Had to test sensitivity as well
    • Even at higher CTs, can get the whole genome with the random nonamer protocol
    • Percentage of reads (coverage) on MinION match Illumina and MinION
  • What about when you don't know what is there?
    • When you can assemble 80% of the genome, it's probably there
    • Three different kits, consistent results

Taking it into the field

  • Testing on Lassa virus, a very divergent RNA virus
  • Lassa is endemic in Nigeria, outbreaks twice a year
  • Take positive samples for sequencing
  • Working great until Nigeria hit more cases than were ever before seen
  • Got data out as soon as it was generated

Processing

  • Only doing DNAse treatment
  • PoreChop
  • Using Canu to assemble, then BLAST to find the closest reference
  • Used De-novo approach to feed mapping

Diagnostics

  • Existing in-country diagnosis uses two RTPCRs
  • Looked at the phylogeny; MinION assemblies were all spread around, all independent spillovers
  • MinION is definitely accurate enough to tell if there's human to human transmission
  • There were two samples that looked marvelously identical
    • Looked at patient forms, found out that they were from the same patient, taken 6 hours apart, one SNP difference
  • NCDC and WHO had a great need to know what was happening; the country as well...
  • Country took the time and attention to listen

Challenges

  • MinION overheated in the first run
  • Time scale changed massively, changed from pilot run to something else

Outcome

  • Were able to make 35 lassa virus sequences in a month