24 min read

Nanopore Community Meeting 2021

[pre-roll music RATM: Killing in The Name]

Update from Oxford nanopore Technologies

Video

Novel statements selected by David Eccles

  • P2 Solo (2x PromethION flow cells): $10,500
  • New Kit 11 cDNA sequencing kits have UMIs and primer fixes
  • Rapid sequencing Kit 11: 25-50ng input (unamplified) on Flongle
  • Short fragment mode coming to MinKNOW: short reads with high accuracy
  • Multi-pass basecaller (Remora); first pass for base sequence, second pass for modifications
  • Liquid biopsy devices that use prep-free electrophoresis for transferring molecules to the pores
  • Browser-based basecaller, uses WebAssembly to interface with GPU

Connie O'Donnell

  • Check out the product tables
  • Add your photo to the virtual audience
  • pre-recorded update from Clive Brown

Clive Brown

  • Second talk of the year, tech should be available either immediately, or in a few months
  • Some forward-looking info on products (disclaimer: research activities, reserve the right to change as we see fit)

Introduction / Recap

  • Goal of nanopore: analysis of anything by anyone, anywhere. Still a lot more to do.
  • Have a membrane, not lipid (an artificial material, very robust). Sits over a well, with common electrode
  • Membrane acts as a resistor
  • Disruptions in the current are decoded into basecalls so that people can make sense of the data
  • Current pore R9.4.1; had R10.3 for a while
  • Motors versioned as E (e.g. E8)
  • Can vary run conditions, temperature, salt; very open to tweaking / modification
  • Downside: takes quite a lot of time and effort to explore that space

Technology

  • need to get stuff out of biology
  • At the moment, need to attach some chemistry on to the front of the molecule
  • Flow cells made in a factory, an array of soap bubbles that sits on a custom-designed sensor chip
  • Nanopore's input chemistry added on top
  • All comes into devices that have very few moving parts, based on electronic design
  • Pretty much everything is designed and made by Nanopore
  • Signal decoding - a large amount of effort gone into decoding

Previous talks

  • Talks earlier in the year; available on the company website
  • Others will go into more detail on the specifics of areas of technology development
  • I get to do a high-level, focusing on my own projects

Read Length

  • Nanopore sequencing samples from solution, distribution of fragments
  • Pores will sample molecules, in almost all cases will read as given; read length = fragment length
  • If you can present the array with very big fragments, will get very big reads
  • Incumbent / market is currently based on short reads; we focused on longer fragment, longer read optimisations and applications
  • Most of the customers have bought based on long reads
  • But... have always been able to do short reads; pore can process down to 20-30 bases, very high capture rate for short fragments
  • Generally speaking, the data is the same. Scaling of signal is different for short vs long fragment, but we haven't pushed that

Short fragment mode

  • A few software tweaks required to optimise short fragment reading
  • Calling this "short fragment mode"
  • If you want to shear the hell out of the library, that will be fully-enabled in the new year
  • Just a software thing to do with scaling; need more files, mostly a configuration thing
  • Then you can do a fuckton of fragments
  • Have been running a short fragment mode on a PromethION chip - 250M native short reads from a human; gives a massive dynamic range
  • No change to the platform, can be run alongside long reads
  • All core benefits, all of it's there on short fragments
  • Short fragment outputs are competitive with the market leader
  • MinION >20M reads; PromethION >160M reads

Accuracy

  • Spoke about earlier in the year
  • An inverse correlation with how much people bang on about it and how much they know about it
  • Steady progress over the last year

Signal

  • Very information rich; extracting value from the signal is the key
  • Work done by nanopore and academics showing that a single atom change gives a detectable shift
  • Latest callers are exploiting the latest advances in neural networks; machine learning team working with external collaborators
  • Have improved ability to call bases and modifications in software only
  • Ability to call non-canonical bases has become interesting

Q20+ Chemistry

  • Chemistry + software; changing chemistry needs optimised software
  • Most people run older R9.4.1 nanopore, and older E8.0 motor; encapsulated by Kit 10
  • Round about now, on pore 9 with kit 10, should probably get 98.7/98.9 averaged [modal] raw read accuracy
  • More recently have launched 10.4, with a modified motor E8.10 (Kit 12); now well above q20 modal. Getting up to Q30 in some cases
  • Still simplex: one pass, one strand
  • Have refined the movement of the motor, a bit slower, not as quick with old chemistry
  • A significant shift in the baseline accuracy
  • Has been out for a while, as part of early-access program, should be generally available on the shop as of Wednesday

Where are we now?

  • Kit 12 is the latest and greatest, beating q20, may still be some headroom in sofware
  • Getting around q30 modal on duplex (both strands on one molecule)
  • Now seeing very competitive consensus quality, particularly on 10/12, seeing q50 bacterial genomes
  • 10/12 chemistry needs lower coverage: R10.4/Kit 12 outperforms R9/Kit 10 on pretty much everything
  • You win, even with reduced output
  • When talking about accuracy, averaging over each error
  • Vast majority of errors is long homopolymers
  • What really matters is the ability to identify molecules, and the ability to call mutations
  • Applications group talking about mutation detecting quality
  • Can run tests with very high confidence (low false positive, low false negative)

Accuracy improvement

  • Continue to improve, improvements in training set
  • Using latest Machine Learning algorithms; adds in to a new loop, each loop gives more improvements
  • Latests kits + latest base callers
  • Research caller (bonito); all latest models are in the caller
  • Some significant improvements in the speed of Bonito; most recent version on par with Guppy
  • Switches on Bonito that will output multiple different file formats (including unmappedBAM), should make re-calling old data easier
  • Nearly a halving of error, quite significant
  • now four different models; most people use SUP - super accuracy; most people are prepared to wait
  • In the near future, this will be simplified; SUp accuracy is q20+ (with 10/12)

Duplex

  • Feature introduced earlier in the year, some people have reported on it
  • Correctly-formed molecule with adapters on both ends
  • Catch one strand and sequence
  • Various tricks to persuade the second molecule to follow the first
  • Can be called individually, but also jointly-called
  • Happens naturally, duplex rate on latest chemistry is about 5%
  • We've had about 50% internally with optimisations ** Have to have adapters on both ends ** Any nicks in DNA will prematurely terminate the molecule

Duplex kit

  • Kit to be released in January that will increase duplex rate to 40%
  • Fragment length doesn't really affect the duplex rate; no strong fragment-size dependency, can access the full range of fragment sizes
  • what you put in = what you get out applies to both simplex and duplex
  • Works fine on R9.4.1 and R10.4
  • Some people running low-yield version of it now
  • Moving to higher yield later on

Calling both strands

  • Work in progress; duplex base caller being developed
  • A lot of effort into simplex caller, now circling back to duplex in accuracy and speed
  • Getting nice q30 accuracy, expected to improve
  • A version of this in Guppy, an older version in Bonito (slower)
  • [skipped]
  • Hard to imagine that q30 duplex wouldn't have an impact on assembly

Methylation accuracy

  • Showed data at AGBT12
  • Underlying problem is a ground-truth problem, need something to learn from
  • Solved using a variety of methods, mostly around synthetic DNA and modified references
  • Really chasing down modified bases on DNA, will then circle back and have another go at decoding direct RNA, where there are many more modifications
  • DNA is the one to do in the short term
  • We can get what we believe is better than bisulphite results; viewed, rightly or wrongly, as the standard
  • R10.4 on Kit 12 outperforms R9 on Kit 10 on modified basecalling

Making methylation accessible

  • All work done my machine learning group has now been reshaped into a slicker Neural Netowrk Method
  • Relatively lightweight network, fast to run, runs as a second step in the basecaller
  • A second pass to annotate modifications
  • Very biologically-plausible-looking molecules
  • Technology, called Remora, is now available in Bonito
  • A switch that will let you choose your output in downstream packages and viewers
  • There's a hit-list of favourite mods: 6mA, 8Oxo-G (DNA left in the fridge), dU
  • Added as the natural process of basecalling
  • Later on will be rolled out to other basecallers, and eventually MinKNOW

Improving accuracy further

  • Still a lot of headroom in accuracy
  • Movement of DNA - a lot of things to play with
  • Large R&D pipeline to keep accuracy improving, all without changing the hardware
  • Duplex main thing is to get yield higher

Longer nanopores

  • Main innovations is to make the length of the pore longer
  • Measure n nucleotides n times
  • Convolve the signal, then deconvolve it
  • Have a pipeline of longer pores
  • Longer pores span longer homopolymers; will call longer and longer over time; average accuracy will get better
  • I just don't like having that chink in our armour

Accuracy and speed

  • The new chemistry speed has dropped a bit
  • started nanopore sequencing at 15 bases per second, found could go up to 450 and still maintain accuracy
  • Latest chemistry has been dialed back to 250 bases per second
  • Speed can be moved around using temperature; has a small effect on accuracy
  • There is a future, focused on R10 chemistry, with run settings on the machine to move the yield around
  • Can think about phasing out the old stuff more aggressively, having a much simpler product

Movement schemes - Foward-looking stuff

  • Going back to the early days... two different ways of controlling nanopore movement: pushing it in, pulling it out
  • Inny vs Outy (like a belly button)
  • Originally wanted Inny, felt more free-running, expected to be high throughput
  • [skip; cat caught a lizard]
  • A lot of time searching the place
  • Want to move back to Outy, a research program to look at Outy
  • Given events over the summer, have refocused the research program on something else

AGBT 2021

  • Slides on F1000 years ago, described Inny and Outy schemes, and pros and cons
  • Some of earliest data was the Outy method

Nanopore Chemistries

  • Another functioning chemistry
  • Motor on one end; distal end
  • Pore sucks in the DNA and strips off (partially) the complementary strand
  • Then the motor sits on the pore
  • Then it will start to move, drags the DNA out of the pore
  • Have it working now at 200 bases per second, want to be 200+ for throughput
  • When it gets to the end it will stop and sit there, giving a stalled / wait state
  • Can then make a decision: eject or keep?
  • Original idea was to use it for sizing the strand

Re-reading

  • Can also re-read a single molecule again and again and again and again
  • Catch DNA, motor out
  • When it gets to the pause state at the end, can open the motor, DNA will zip back through
  • Can shuttle backwards and forwards on one molecule to a very large number of iterations
  • Even with fairly naive sofware, averaged basecalls, single-molecule accuracy gets better and better
  • Unoptimised chemistry, get high accuracy consensus on a single molecule
  • Under software accuracy, so real-time control over doing this; how many copies you want
  • Can be basecalling and assessing until you hit the accuracy you want
  • What really matters is your ability to definitively call a mutation or modification at high confidence
  • Need to be confident beyond q50 [not there yet, around q30 now]

Adaptive sampling

  • Can taste sequences; can scan them
  • e.g. molecule from an oncogene, or river water
  • Can fish for interesting molecules in one step
  • Once an interesting molecule has been found, can accumulate on the pore
  • Can go up and down it until we've got the accuracy we need
  • Read until the aggregate basecalling confidence is definitive
  • Adaptive sampling until mutation confidence crosses a threshold of significance
  • Looking for a rare molecule, a needle in a haystack. Once you find it, need to be confident about it
  • Accumulate re-reads as required for the application, very high accuracy single-molecule consensus

Why do this?

  • General tissue background
  • Some cells have rare variants; thing you're looking for might be 1 in 1000, or 1 in 100,000
  • In our scheme, can taste and ignore uninteresting molecules
  • Frees up time for wasted capacity in the sequencer
  • Can find needles in haystacks, and look at them, and tell who made them - the essence of liquid biopsy
  • River, blood, air droplets
  • Averaging over multiple observations of one molecule, because those molecules are rare

Applications

  • Blood, subways, rivers, anything where there's a heterogeneous mixture of fragments where looking for a subset at high confidence
  • Laboratory-based workflow is relatively straightforward
  • All on the same hardware, don't need to make a new box
  • I think we can get to that fairly quickly (middle of next year)
  • Much more difficult is doing this in a way that doesn't need a lab

Liquid biopsy devices

  • Liquid biopsy platform - bottom chip, cis chamber with DNA, sample chamber above that, pore matrix above them
  • Can move DNA molecules (and all molecules) through electrophoretically
  • Geared up towards swab inputs that don't require pipettes
  • Gives point-of-care, point-of-field device
  • When electrophorese things through the chamber, the first things that go through are small charged molecules
  • Feasible to also think about small molecule sensing on this platform
  • Second thing that goes through the membrane is small DNA
  • Moving towards a reservoir on the chip, no external reagents, a device that lets you flush with a button press
  • Can put through strands, can spike things into blood
  • Still a lot of work to do with this, very early days

P2

  • Mentioned a while back, devices should be shipping in mid Q2 of next year
  • 2 PromethION flow cells with integrated computing
  • Price $59,955, will allow human WGS for under 1000
  • Between GridION and PromethION
  • Low to moderate flow-cell usage, multiplexing enabled, whole plant genomes

P2-solo

  • Easier to make, will be out early Q2 (maybe earlier)
  • 2 PromethION cells, max R9 output is 500-600 Gb per run
  • Also working on parallel-pipettable multi-channel flow cells (not available Q2)

Browser calling

  • https://bonito.epi2me.io
  • Want to go back to remote calling
  • Java vs WebAssembly
  • Latest browsers can access the GPU on your computer; a way to deliver low-configuration software to people
  • Web Bonito, runs entirely within a web page
  • All data is local. If your computer has a GPU, it's all working quite nicely
  • Pretty good, given that it's in beta at the moment
  • Don't need to be a bioinformatician to re-call data
  • Drag and drop the files, bring them all up-to-date

[skip by ONT]

Mk1D

  • Intend to replace Mk1C with Mk1D
  • Based on iPad Pro
  • A custom case based on MinION sequencer
  • Battery, can run offline
  • Can exploit all the features on the ipad pro
  • Web-based Bonito can be run on M1 processors
  • Getting keep-up basecalling, 100 kb/s, on iPad Pro
  • Getting close to P2 calling on M1 Macs
  • Looking at very late next year
  • Heat control is better
  • Thinking about an area for other devices (e.g. sample prep device)

Round-up headlines

  • Raw reads now q20+
  • Duplex q29
  • Assemblies extremely good - would like to see a duplex ultra-long assembly
  • Integration of modified basecalling in existing basecallers
  • New product coming out (P2)
  • Would like to see good demonstrations of liquid biopsy

Org.one

  • We help and subsidise the sequencing of reference genomes
  • Prefer if it can be done in-country
  • [skip; cat caught another lizard]
  • using eDNA to have positive outcomes (tracking / tracing, improving outcomes)

James Clarke

  • Nanopore goal: analysis of anything by anyone, anywhere
  • Look forward to seeing what's coming down the pipeline
  • Accuracy is a topic that is discussed a lot
  • Went out the door very early, knew accuracy would improve, have come in leaps and bounds, probably exceeded the expectations of a lot of people
  • Changes can perhaps be a bit confusing
  • Talk about our technology's headroom
  • Will concentrate on what we have right now; single molecule accuracy with over 99% confidence
  • Can't think of any other analytical technique with that accuracy
  • q50 consensus accuracy

Real-world examples

  • Can aquire a vast amount of data from one instrument
  • Signal is very information rich, just need to decode them
  • Long reads have been a big draw, incredibly large genome [43 Gb]
  • Very quick library preparation step, able to take your sequence to your sample

Previous talks

  • Can go to nanoporetech.com to see other talks and content

Nanopore technology

  • Have to get DNA from our sample, adding in adapters that contain the molecular motor
  • Adapter loaded into a flow cell
  • Strand moved through nanopore by an enzyme
  • ASIC moves signal to basecaller
  • Array of many nanopores, work asynchronously, a very dynamic system

R&D

  • Accuracy: 3 main things: nanopores, enzyme (as consistent as possible), decoder / basecaller
  • Other things that can affect accuracy: temperature, salt, voltage, capacitance from membranes, ASIC digitisation, instrument noise
  • Output is about the number of nanopores you have, determined by ASIC
  • Pore yields have been improving steadily
  • About a third of time is spent in lower levels

Instruments

  • PromethION - 48 flow cells, each up to 3000 channels at once
  • Flongle - 126 channels, cheapest

Platform Update

  • Accuracy - moving through the company and out into flow cells
  • Slight change from 10.3 to 10.4, slightly more accurate, easier to handle and get into flow cells in high numbers

PromethION

  • board throughtput is limiting, upgrades pending
  • Incredibly high-throughput device; one run with 48 pores prouced 10 Tb of data
  • Can run in different modes
  • NovaGene now have largest field record from a single flow cell: 245 Gb
  • MinION has slightly bigger wells than PromethION
  • Mediator underneath each channel, gets consumed over time. MinKNOW does this, but ideally we don't want to do this
  • PromethION well size has been increased, so reduces mediator compensation needed
  • Early patches have fewer active pores, but more yield, getting 270 Gb from PromethION flow cell

PromethION P2

  • Should be really popular
  • Leverages the cheap cost-per-Gb from PromethION
  • Runs 2 PromethION flow cells
  • Two configurations, one with compute built in, one without compute device
  • P2 solo can be connected to a GridION for calling

Flongle

  • Pretty simple concept - put expensive components into the instrument
  • Slight tradeoff - need to connect the flow cells to the ASIC, 126 channels
  • Community record at 2.83 Gb, better than internal data
  • Constantly debugging Flongle, need to be very careful with power management
  • When charge is built up too quickly, membrane can pop
  • For MinION, ASIC manages power control
  • For Flongle, need to be more careful and considered
  • Software update, protocol change

PCB Flongle

  • Point of flongle: make flow cell as cheap as possible
  • Alternatives already used in the industry
  • Proof-of-concept with 100 channels, got 1 Gb
  • Team challenged to get 126 channels, got to 120 (to do with manufacturing challenges)
  • Happy with 120
  • Chips are backwards compatible, utilises learnings from PromethION volume changes
  • Have manged up to 2 Gb

Flongle Prototyping

  • Want to remove the need for you to flush your flow cell
  • Flush is there because we are using Pt electrode
  • Looking at moving to silver
  • Will probably test on Flongle first, MinION in the future

ASIC

  • Does a number of things, defines max channels
  • Takes chemical signal and digitises
  • Perfoms a number of other functions: voltage control, flicking
  • Generates a lot of power; needs to be dissipated
  • New ASIC with 400 channels, very low power
  • Low power means instruments can be very simple, use power from USB devices

New ASIC sequencing

  • Breadboard with new ASIC attached
  • Attached standard Flongle flow cell to ASIC
  • The first look at instrument develop
  • If ASIC is the engine of the platform, this is like strapping an engine to a go-kart
  • Isn't even running at the right temperature
  • However, getting characteristic squiggle, accuracies of the run are pretty respectable [modal 97%]

Stuart Reid

  • Will talk primarily about accuracy
  • Innovation / change / disruption
  • Continuous improvement process
  • Major step changes are run through a developer program, looking for a collaborative workflow
  • Eventually moved through with lead times, warranties, notifications

Accuracy

  • Raw read / simplex, now beating q20
  • Duplex accuracy approaching q30
  • Seeing q50 bacterial genome with ~20X coverage
  • [skip for emails]
  • Platform is sensitive, high-performance

Changes

  • Improvement from algorithms alone, over 2 years accuracy improved from 92% to over 98%
  • Step-change improvements to chemistry to change the signal; disruptive to some degree
  • Changes to the chemicals, not the sequencer
  • Latest chemistries beating q20 raw read accuracies

Basecallers

  • Fast callers keep up with sequencing on all devices
  • Super-accuracy basecallers are right at the top end

Duplex technology

  • By reading both strands, can get up to a modal accuracy of almost q30
  • Run both signals through the basecallers, caller creates a call that is consistent with both signals
  • Present in Guppy, will be flipped to a full release, then MinKNOW

Consensus accuracy

  • Consensus erorrs average out with coverage
  • Recommend Medaka tool for polishing
  • Best results with R10.4 and Kit 12
  • Kit 9/10/11 still has pretty respectable results

SNP and INDEL calling

  • Making a change to recommendation
  • Recommending Clair3, replaces Medaka variant caller
  • Fully-supported EPI2ME workflow
  • A few features that we were considering implementing in Medaka were already implemented in Clair3

Variant calling results

  • Robust all-round performance across both chemistries
  • Can get all forms of variation from a single experiment
  • SNPs at 99.9% (R10/Kit12)
  • INDELs at 89%
  • Structural variants at 96%
  • Methylation is the best in the business

Methylation calling

  • New way - Remora
  • Results as good, or better than existing techniques
  • R 9.4.1 / Kit 10 was already better than bisulphite
  • R10.4 / Kit 12 results are significantly better than 9/10 results
  • Remora maintains existing accuracy; a second pass through the signal
  • Modifications annotated on top of the existing signal

Results

  • Can buy your own MinION
  • Host open datasets (best entry EPI2ME labs)

Barcoding

  • Nanopore barcoding is accurate
  • Single-ended barcodes recover over 90% of data
  • Barcodes on both ends get 99.99% correct

Plants

  • Can do plants, but accuracy is a bit lower
  • First point of call in error is to PCR sample, points towards modifications in error
  • Have added plant modifications to traditional model, want to call the canonical sequence

Short Fragments

  • Nanopores have always done short fragments
  • Have a lower-limit on the output reads
  • New configuration, write out the short reads down to 20 bases, built into MinKNOW as the new default
  • Over a quarter of a billion short reads, around 25% perfect

Accuracy summary

  • Duplex getting on for q30
  • Assemblies are highly contiguous, highly accurate
  • Circular bacterial genomes q50
  • Methylation for free
  • Highly-sensitive test records

Software

  • MinKNOW - the sequencing software that drives your sequencing
  • MinKNOW is designed to be point-and-click, has expert settings just below the surface
  • Can take action as soon as you have the information you need
  • Have a large stack of work in progress, various optimisations, integrated bioinformatics, increasing focus on user experience

Portable

  • Portable devices can be taken to the field
  • App for Android devices
  • Basecalling is getting faster, fast mode keeps up on all devices
  • Fast basecallers are getting more accurate
  • Also expanding general support for GPU integration, automatic GPU detection in the new year
  • A stack of speed improvements to come in

Adaptive sampling

  • Unique to ONT sequencing
  • A huge range of things that can be done
  • Barcode balancing in progress
  • Up to 10-fold improvement in target
  • First-class API that is fully-supported

Rosemary Sinclair Dokos

  • Talking about packaging of the innovation
  • Scalable platform, highly accessible, very versatile, information rich, high accuracy, high output, real time
  • Started from Flongle all the way up to PromethION P40 (10 Tb)
  • This year have launched into relationship with Eventor
  • Have now put P2 onto shop

Flow cells

  • Incredibly transparent
  • Adding 12 pack to Flongle
  • Adding 96 pack to MinION
  • Price per Gb - incredibly competitive
  • Flongle $90 / Gb, MinION $60 - $10 / Gb; PromethION $10 - $2 / Gb

Versatility

  • A lot of levers to play with
  • Can choose run time
  • Can choose chemistry speed
  • Can choose fragment length
  • Can multiplex, run different algorithms
  • Versatile platform because customers have versatile demands
  • Can track Twitter, see what other people are doing

Information

  • Methylation, SV, assembly
  • Get all information irrespective of the type of experiment
  • Accuracy performance doesn't need much depth (methylation high accuracy at 20X)
  • Phasing 98% of human genome
  • Applications team work with users to bring metrics to life
  • Complex genetic disorders, tech lets users untangle that

Rapid, real time

  • Reads are immediately available for analysis, allows rapid answers in the field

Time

  • Products have a different time / bounds to operate
  • Developer -> Early access -> Released -> Fully released
  • Started with MinION in 2014, added GridION, PromethION, Flongle
  • Driven by upgrades in software, chemistry, consumables

Enhancements

  • Accuracies (97% a year ago, pushed to 99% now)
  • Now have reports of users in the field with over 50Gb from a single flow cells
  • Incredibly clean / crisp data

Whole New Platform upgrade

  • Shifting up a gear into the 99% field
  • Some established chemistries, established motor
  • Will continue to support users undergoing current projects
  • New high-capture adapter
  • PromethION M chip
  • q20 specific upgrades ** New Nanopore ** New Adapter ** Duplex ** PromethION device upgrade
  • New baseline: R10, Kit 12

Products

  • Preparation - potentials with lower inputs
  • Rapid kits from 50-100ng, looking at pushing across the whole range

Top kits in the process of releasing

  • Native barcoding for q20: 24 kit $700 (6 runs); 96 $800
  • Sequencing kits with integrated barcoding
  • Rapid kits being upgraded to q20
  • Reduced input
  • Fixed an enormous amount of things to make it Flongle-friendly
  • 25-50ng for Flongle

Multiplex kit

  • Validation from customer sites
  • Target: 30X coverage per genome, looking at two genomes per flow cell

Ultra-long sequencing

  • Ability to go into megabase reads
  • NEB Monarch kit (cell, blood, tissue); prep in about a day
  • Short read exclusion kit, removes short reads before sequencing

Adaptive sampling

  • Great capability for Oxford Nanopore
  • Launching reduced-representation methylation sequencing; comparing MinION flow cell vs bisulphite, superior performance

cDNA

  • Pairing up fuel-fix adapter with cDNA kit
  • Have generated 20M reads from MinION, 160M of PromethION
  • Included PolyA length determination
  • Incorporated UMIs
  • Live, in stock today
  • Need to do artefact depletion
  • Looking at single-cell sequencing via PCS kit
  • Longer read, with expression

Midnight ARTIC sequencing kit

  • Very simple prep, about 5.5h, strong performance even at high CTs
  • Can get going from $10 per sample

Short sequencing

  • Short fragment mode
  • Now have short read chemistry with high accuracy, and methylation, accessibility, portability, real-time
  • GridION - could be running short on one flow cell, ultra-long on another

VolTRAX

  • PCR-enabled devices currently in the air, look forward to more talks about that in the new year

Sequencing

  • Great, fantastic range of products
  • MinION - $10-$25 per Gb, can sequence hundreds of flow cells per year, can do small, targeted, Cas9, so many different applications possible
  • PromethION - P2 can now sequence 1-200 flow cells per year (72h); P40 for really high output

Population-scale genomics

  • Dedicated team ready to help with automation setup
  • Some users running single genomes, pushing into 2 genomes per flow cell
  • Simple prep, single Hamilton unit can fill 6 PromethIONs, still Plug-&-Play

PromethION upgrades

  • Marathon chips [with larger reservoir]
  • Hitting the streets in early 2022
  • Orders will migrate towards this; it will be the new norm
  • Working on A100 towers to increase GPU / CPU / RAM / IO
  • P24/P48 tower upgrade

P2 Solo

  • Starter pack from $10.5k
  • Fits nicely along GridION
  • Have reconfigured pricing / packs
  • Human genomes for under $1000

Analysis

  • Everything has moved on since London Calling
  • Getting everything moving smoother, quicker, better
  • Remora - will be able to get methylation at the same time as basecalling
  • Teams are working on bringing this all together - Just want to load your data and the files you want to see come out of the end
  • Ability to load sample, MinKNOW demultiplex, Phase / call SNPs, methylation -> BAM/VCF file and FASTQs

Interpretation

  • Analysis pipelines
  • Research software released on GitHub
  • NextFlow / EPI2ME lab workflows; really clean deployment
  • For analysis on the fly, have regionalised EPI2ME to improve connetivity and speed

Summary

  • Incredibly scalable platform
  • Accuracy has moved enormously over the last 12 months

Release timeline

  • VolTRAX out December
  • P24/P48 upgrades; tower upgrades Q2/Q3 next year
  • Right now native barcoding kit for Q20, PCS kits, single cell protocol
  • Software side: Remora
  • Feature request pinboard
  • Focus to make product as simple to use as possible

Q&A

  • Will short fragment mode work with RNA? Not yet. Actively working in the background.
  • Protein sequencing? Working on it, long-standing relationships with academics, not quite ready for prime time.
  • Adaptive sampling? Nothing released for PromethION, requires firmware update behind the scenes.
  • Will old kits work on R10.4? Currently supported for Kit 12, working on compatibility; different configs to run.
  • Will R10.4 / q20 need refuelling? No.
  • Accuracy - will it vary according to the sample type? Shouldn't vary, but it may well do. For example, base modifications in plant DNA; some interesting science to be done there.
  • Would R10.4 be recommended for all DNA? More accurate in most cases, but still some reasons for R9. We will keep supporting R9
  • Duplex vs 1D^2 kit? 1D^2 kit is a chemistry that is no longer supplied, worked in a similar way; required modifying the pore, adding additional sequences to the kit. Duplex is more efficient now, have enhanced the natural tendancy of strands to follow each other. Other strand probably returns back into the solution, a small chance it gets back into the pore.
  • 5mC vs 5hmC? Yes, can distinguish, looks biologically plausible.
  • Does web calling need NVIDIA GPU? No, will take all sorts of GPUs.
  • Next plans for increasing basecalling speed Performance engineers are getting through the list, a bit too early to make the cut. Will be added to platforms.
  • What steps can be taken to maximise yield? Many resources on community, master classes. Sample prep, extraction. Tech support, field application specialists.
  • Org.one expanding past red list We're all incredibly passionate about it. Have started with red list, were pushing for proof-of-concept, wanted the ability to generate data anywhere. Website coming out shortly to select breeds and place bid. Other species can be submitted, as people are interested.