London Calling 2017

Presentation notes on London Calling, written by David Eccles.

Day 1 (May the Fourth be with You)

Gordon Sanghera

The Kinks (intro sound / theme sound for this conference)

• A lot to do with nanopore sequencing
• Disruptive; appeared between The Beatles and [some other band that I didn’t have time to take note of]
• Used knitting needles to slash speakers, basically launched heavy metal and punk

Lord Kelvin

• Without measurement, there is no scientific progress
• We are now taking measurements and gathering information in real time
• Real time changes everything

Billingsgate (location of conference)

• Named from a water gate where goods landed. It was managed by King Billing
• In 1699 an act of parliament allowed a fish market in Billingsgate, no other market was allowed within 6 2/3 miles
• Except for eels – they were allowed to be sold by the Dutch, because of the help the Dutch provided during the Fire of London
• Yesterday, a Dutchman bearing eels gave a presentation
  • Jellied eels are a local delicacy; attendees should try them out

Disruptive Innovation

• Venue has been organised to seat 400 people; it has filled up
• There will be a guest speaker at dinner talking about disruptive innovation
• It is you lot [i.e. the attendees] doing the innovating
• Some of us have gambled our whole careers on this
• ONT is here today to learn from the community
• 25% of ONT staff are here to listen
• We are here at the beginning
• Many careers will be made through the bravery of the community

Zoe McDougall – Housekeeping

• Packed program; any speakers should state at the beginning of their talk if they don’t want something tweeted
• WiFi available; Battery charging points all over the place
• No scheduled fire drills, so if you hear an alarm, leave the building through the exists [DE: there was actually a scheduled drill at a previous London Calling conference]
• Three floors; eating and drinking will be done tonight in the basement

Karen Miga

The Use of Really Long Reads

• New excitement in the genomics field for reads 10-100kb in length
• Still not completely there with the longest reference assemblies
• Huge gaps exist in the human genome
• Most sequencing technologies take place on the chromosome arms
• It is difficult to get true haploresolution on the centromeres
• Centromere repeats are head to tail for millions of bases
• A fundamental genomic milestone is almost within reach: telomere-to-telomere assembly

Assembly problems

• Repeats are 98-100% similar
• A short library with reads of 10kb can need reads of up to 100kb to span
• With human centromeres, even 100kb reads are too short
• What about unique markers within the region?
  • There are some scars and SNVs
  • SNVs are probably the most prevalent thing for assembling
• One sequence is not the problem
  • Diploid read phasing is incredibly difficult
• There is sequence similarity for different genomic regions: satellites on chromosome 18 share with chromosome 20

Resolving the issues; 3 key advances required

• Understanding of satellite sequence structure
• Increased throughput
• High-quality base calls

Satellite array on chromosome Y

• Why Y? Y chromosome has the smallest satellite array, and is well characterised
• Have used a BAC-based strategy to assemble
• There are a set of 9 BACs that are known to span the region
• Assembly process: sequence BAC to high depth, use Illumina reads as a truth set

Using UCSC Longboard protocol

• Linearising process for BAC, optimised to give the BAC sequence
• Managed to get a 100kb+ [library] N50
• In total, there were 3.5k reads of greater than 150kb
• From the nanopore reads alone, the accuracy was not high enough for centromeric analysis
  • Polishing can be done to end up with high-quality consensus
  • Polishing works well with about 10~60X coverage

BAC sequencing example: 221.4kb

• Longest sequenced microsatellite region
• Can now detect 634 errors in this region
  • Lots of the errors are homopolymers A & T
• Resequencing carried out on Illumina, including pentamer analysis
  • Comparison with Illumina gives strong correlation
  • ONT had only one outlier, AAAAA/TTTTT
• After error correction, only 23 of the variants seemed to be true

Other BACs

• All BACs put together produce 346kb centromeric regions
• Assembled to R1B [particular human reference], median repeat region 350kb, SD 220kb ~ 460kb
• Looking at the individual cell lines, can find restriction enzymes that do not cut within the array and haplogroup-matched dataset
  • get blot at the expected region

Future work

• Move along from BAC-based approach
• With long reads, can probably span these regions
• Optimising the 1D protocol, ideally want to get to 60X coverage for long reads
• UCSC working on visualisation for variants and centromeric regions

Questions

• How long to reads need to be to do all centromeres?
  • A megabase sized read might work; the mean centromeric length is 3Mb
  • If 3Mb can be spanned, then it will be possible to do a diploid resolution of centromeres
• Why does accuracy go down when increasing coverage?
  • False positive consensus; there is a sweet spot

Björn Usadel - Bringing Omics Data to Users

Plant Genomes: a historical perspective

• Plants are particularly nasty beasts with very large genomes
• Plants are riddled with repetitive sequence
• Wheat is particularly nasty (12GB)
• The whole genome and gene space is repetitive
• Multiple copies, even for small genome families

Tomato

• Contrary to the usual definition of species, different tomato species are all very happy to cross with each other
• Solanum penelli is a very hardy plant
• Did sequencing with Illumina, fosmid, BAC
• Got scaffold N50 of 1.6Mb, contig N50 of 45kb
• Sequenced a new “cultivar”, but there are millions of variants

Oxford Nanopore and Plant Genomes

• Started with small algal genomes
• Canu and Pilon polishing worked well
• Processed 31 flow cells, each with over 1Gb per flow cell
• Read length was tunable, but a little bit dependant on what can be selected with the Pippin Prep
• Output quite high in 24 hours
  • After the run was done, the spent flow cell could bee used for testing purposes
• Fairly good correlation between reported q value and actual quality
  • 81-82% accuracy
• Very clean DNA is needed; herbal plants can be problematic
  • Read correction with Canu improved accuracy to about 90%

Assembly experimentation

• Tried a whole bunch of assemblers, tried subsampling data
• Started with 100X coverage
• Canu correction together with SMART-denovo got the best N50
  • Canu makes fewer mistakes; Canu-corrected reads are fed into SMART-denovo
• Is SMART-denovo the best to use?
  • It was much faster; Canu-corrected took 10,000 CPU hours
• Tried sampling, putting emphasis on getting long reads
  • 30X coverage with longest reads works best
  • In general, looked good when mapped to the reference genome

Other quality metric: BUSCO

• All methods had similar unpolished BUSCO scores
• Pilon polishing worked the best

S. penelli Compared to other genomes

• Pineapple N50 120kb
• Quinoa scaffold assembly 3.84Mb
• Citrus PacBio + Illumina, contig N50 2.2Mb, scaffold 4.2Mb
• Can actually get a good assembly in a few months

Future / Questions

• Finishing other genomes
• Waiting on albacore improvements: longer insertions, longer deletions, overall better
• With R9.5, got one read that aligned well to chloroplasts
• What does it mean? Small labs can sequence a genome
• Repetitive elements are being analysed now. Centromeres are not being talked about
• Does Pilon introduce noise in the repeat regions?
  • After 5-10 rounds of Pilon, things started jumping around
  • Prefers 2-3 rounds of error correction with Pilon

Jared Simpson – Analysis Tools For Nanopore Data

Signal-level nanopore data

• Want to work with raw signal as much as possible
• Channel passes single-stranded DNA through a nanopore
• Current samples are written out to FAST5 files
• Sequence disrupts the current
• DNA movement introduces a new sequence context
• We hope to see movements up and down that reflect movement of the DNA through the pore
• Signal-level analysis involves working with both raw samples and segmented currents (events)
  • Events are fewer in number
  • Event-based algorithms typically run faster, not a lot of data lost

Basecalling

• The primary analysis task of nanopore
• Involves inferring the sequence that gives rise to the current
• Take a vector of events, apply labels to the events
• Overlapping labels are merged together to give a final base-called sequence
• Don’t want any loss of information in going from events to base calls

Event data / Nanopolish

• Larger files, slower models
• Nonetheless, better improvements in data
• Nanopolish was originally for improving consensus sequences
• Can now call SNPs, INDELs, and modified bases
• Can also be used for phasing reads and long-range haplotypes

Lightning Talk – Raja Mugasimangalam

Is The Pot Labelled Correctly ?

• From India, taking us back several thousand years ago
• No dairy products had labels, all in pots
• Now have labels, but wanted to know if they were correct
• ViBact labelled well; it had a single microbe in it
• Yakult labelling okay, but could be better
  • Found some fungi that shouldn’t be there
• Home curd is unlabelled and untold
  • The process of discovery: stir / spin / extract
  • 12 products + 9 other
  • Could identify pot with a very low number of reads
• Need at least 500 reads for a complex genome (if more than 10 bugs)
• Major composition of samples was determined in 4-6 hours
• Now have local calling
• Were there reagent contamination issues?
  • Used controls with common contaminants
  • Did multiple samples from different shops
  • The same lots contained the same bugs

Lightning Talk – Sebastian Johansson

HLA Typing

• HLA is important, especially for organ transplantation
• There are two HLA classes of varying length
• Got products from an amplification primed on HLA sequences, pooled together
• 24 hours of sequencing produced a good barcode distribution
• Coverage was good for short reads, not so good for long reads
• Looked at alignments to HLA after filtering
  • Got small islands of coverage that disappeared after filtering
• A few reads were found spanning the entire HLA region
• Works well for A, B, C, DQ
• Would be improved with spanning primers

Lightning Talk – Franz Josef Müller

Rapid Identification of Genomic Regions / SelectION

• Developed tool is a solution to a specific problem
• Intermediate step between basecalling and alignment (anchoring)
• Cuts down on the time required to carry out an alignmment
• Localised 160 reads to FMR1 region; 28 reads to repeat within FMR1
  • Subsequent analysis time was less than a minute
• Essentially Burrows-Wheeler Transform + FM Index
• Four orders of magnitude faster than mapping everything ‘ Mapping of over 90% of the reads, accuracy over 90% compared to BWA
• Nanopore sequencing excels at repeat expression finding

Lightning Talk – David Eccles

Sequencing a parasite genome

• A long history of sequencing, started doing his own prep with R9 kit
• Runs in October / November last year were bad
  • ONT offered to do the sequencing for them
• Five sequencing runs, all better than 20 previous runs at institute
• One run amplified from a single worm
  • Metagenomic profile from a 35,000-read subset
• One run had a very long read length “hump”
  • Sonication + Tip20 extraction
  • Run N50 was better than current reference genome
• Poster talks about assembly discoveries

Lightning Talk – Sally James

Telomere to Telomere sequencing of Galdieria sulphuraria

• A real pain to lyse
• Telomere is an octamer repeated 26 times; other end is repeated 29 times
• Nanopore reads aligned to an area right at the ends of contigs
• Reads were piling up at the ends of contigs
  • Many 40bp reads, pileup and completely stop dead at the end of the contig
• Can look at multiple repeats
• Currently assembling complex sub-telomeric regions

Lightning Talk – Beth Lodge (ONT)

VolTRAX

• All electronics are in the base of the unit, flow cells are added on top
• Voltrax integration program is open, applications are available online
• High-output kit, gives best throughput
• Yield variation between labs is reduced significantly
• It can do extraction as well as PCR
  • Similar yields to other extraction methods
• Looking at 16s sequencing
• People don’t need a fully-kitted out lab to do sample prep

Philipp Euskirchen

[DE: starting to feel jet lag problems]

[History of brain tumour diagnosis]

Diagnosis

• Tumours can have very different morphologies but be molecularly similar
• Diagnosis is preferred if it leads to a prognosis
  • e.g. identification of an H1 mutation; a 1p or 19q codeletion strongly correlated with prognosis
  • For subgroups of medulloblastoma, prognosis depends on the subtype
• How can diagnosis be implemented?
  • There are four subtypes of medulloblastoma based on their transcriptomic and methylation classification
  • A lot of different machines and protocols
• Two paradigms
  • Low-pass whole-genome sequencing, producing results within a day
  • Deep amplicon sequencing for point mutations and GC-rich regions

Copy-number profiling

• DNA with rapid kit and sequencing
• Used out-of-the-box tools from the supermarket
• Read numbers are not gigantic, but could still generate decent profiles within a day
• Visually explored the EGFR region on chromosome 7
  • Saw focal amplication
  • Was able to identify split reads at both ends
  • Remapped to a synthetic sequence for collapsed reads
  • Confirmed by sanger sequencing

Methylation sequencing

• Can use the same reads as used for copy-number profiling
• Compared with Illumina 450k chip
• Wanted to know if this could possibly work for nanopore
  • Classifier error changes things; don’t know in advance what CpG sites will be seen
  • Compared with Illumina, methylation profile has a nice correlation

Pan-cancer classification

• Classification from a while different method x * Works for IDH mutant and medulloblastoma
• What works in brain tumours should have benefit for other cancers

Real-time read depth monitoring

• 10-20 minutes until a region is 1000X-covered, including a GC-rich region
• Results mostly in line with Illumina data

Breakout 3 Discussion

• Nanopolish doesn’t work so well with the new version of albacore
• ONT wants to work to produce a better base caller, want to avoid needing nanopolish
  • Don’t want a base caller that introduces systematic error
• Damien has tried tracking virus with mouse samples, seen different diversity in different tissues
  • With direct RNA sequencing, even with 10^6 to 10^7 copies per ml, sill gets swamped out by the host
• What is the optimum depth?
  • Read length matters
  • With good long reads, 20-30X is good enough
• For clinical use, need to be very sure about accuracy
  • There are applications where read-until will be useful

Clive Brown

Novel statements selected by David Eccles

• Thinking of making a GridION version that has PromethION flow cells
• PromethION Will also do a 4-way MUX (max 10,400 wells)
• New chemistry coming which will increase MinION yield by 20-30%
• Accuracy should get to 99.9% [q30] by the end of this year (possibly this summer)
• Updated pore [technology development] has two times read-ahead, and should be able to model up to 30bp homopolymers
• A reasonable user will be able to knit their own clothes using nanopore packaging
• Have been able to get a metal-metal interface working on the Flongle
• A sequencer that doesn’t do base calling – Clive’s pet project for the rest of the year

[evening for Clive brown is after 6pm]

Goal of ONT: Anyone sequencing anything anywhere

• Most products are targeted at this gooal
• Sequencing is becoming more ubiquitous
• When things get out of the lab, people find out more uses
• People have gone out and found real applications
• Can go back and mark Clive’s homewark as a permanent record

Clive explaining nanopore sequencing

• Protein pore at the moment
• Pore will scan the entire sequence length
• Other technologies are typically limited by photo damage
• Sequencing is fast: per-molecule cycle time is milliseconds
• Should have significant implications in the future
• Fragment length is effectively read length
• No reason why full-length chromosomes can’t be done
• Histones shouldn’t hinder sequencing
• What if we could do whole chromosomes?

MinION

• Probably all people at the conference are MinION users Have always
• Felt MiNION would be good enough; it is the democratising feature of ONT
• Sits in no-mans-land because a lot of electronics are in the flow cells
• There are now about 4,500 users, ideally want 10,000-15,000
• Workflow is never linear with the MinION
• Can analyse data in an hour, some people are beginning to do this by not just firing and forgetting it

Lab Work

• Clive is not good at lab work, but has had a go anyway
• Clive has run what ONT has to see how hard it was
• In half an hour got 140µg DNA using a Qiagen kit
• Could load flow cells (got one bubble once, but easily fixed by other ONT staff)
• First attempt used 30 flow cells
• Subsequent run on GridION, 5 flow cells at 14-15 gigabases
• Some users have got 16 gigabases
• Some have trouble with getting anything
• Clive used reagents that we have now, demonstrating that the GridION can be used now
• Another run used newer chemistry (not available yet in the community), and achieved 24 gigabases per flow cell
• Now has over 200 gigabases, people can do with that whatever is ethically reasonable
• Probably some protocol issues; some samples don’t yield much DNA
• People are now using VolTRAX to give consistent yield; flow cell is now tip top

GridION

• Started as a modular rack-mounted system with reagents, dynamically forming membranes and sequencing
• Abandoned this approach, but left it parked on the website
• New GridION is modular
• Thinking of making a version that has PromethION flow cells
• Will have Run Until and Read Until
• People are licensed to use the GridION for commercial use
• Can alternatively use five MinIONs, with laptops, software, etc.
  • Process is all automated on the GridION
• Box can do real-time online basecalling

FPGAs

• Internal effort to map data processing into FPGAs
• Can call 1 million bases per second at the moment
• Not as much success with GPU / CPU
• Gives very fast feedback, even with a high number of pores

Funding model

• ONT doesn’t like the idea of how people pay for things
• Preferred model is consumable only; almost Pay-As-You-Go, with contracted minimum
  • This has been the most popular model

Manufacturing

• Box is very highly manufacturable
• After June, should be able to make 3 per day
• Shipping in about 48 hours
• Flow cells have about a 48 hour delivery
• First GridION out on the 15th May, might be the same GridION used for the Cliveome, but maybe not

PromethION

• Designed 3 years ago, everything is completely new
• Was designed to take on Illumina
• In principle, will generate more data than a NovaSeq
  • Head room is so much that even if NovaSeq is brought forward, will still be competitive
• Have 12-14 boxes out now, produces a ton of data
  • Bulk purchase cost of flow cells is about $600
  • Just sit tight
• Can de-batch the entire workflow
  • e.g. 50 samples on a Wednesday, one sample the next day; more real time and less batched
• PromethION is licensed as fee-for-service
• Flow cell has a different layout
  • A few issues remain, but nothing that hasn’t been seen before with the MinION
  • Even with issues, getting 22 Gb in 22 hours [DE: I recall there was a ‘23’ somewhere]
  • Active unblock is not yet fixed on the PromethION
  • Flow cell run time has been increased to 4 days
  • Will also do a 4-way MUX (max 10,400 wells)
• Even if just using similar output to a MinION, is still better than NovaSeq

PromethION Processing

• Increased speed of sequencing
  • Internally have a run at 1,000 bases per second
• Couldn’t fit processing into the original design
• Solution is to make as separate computer
  • Will eventually shrink by about 1/3
• Composed mostly of FPGAs, will allow complete base call in real time
• Lower module will be a network switch
• Inside is a server with a skin
• FPGAs with neural networks
  • Loads of them out and configured
• Steady incremental improvement

PromethION evolution

• Firmware tweaks are increasing processing to 4.8 Tb
• Ultimately will retransfer computing, end up with a single server
• People should sit tight for this one; this is the box that will turn the supertanker market around

MinION yields

• Customers are getting pretty good yields
• Should be able to get 20 gigabases per day, but more realistically 6 or 7
• Someone got 16 gigabases [one run], Clive wishes they would tweet that run
  • 16 gigabases is a realistic expectation
  • New chemistry coming which will increase that yield by 20-30%

MinKNOW

• Improving rapidly
• Should be able to wash out flow cell sample, fix sample prep, and move on
• Want to get rid of the priming step
  • Just load the flow cell without priming
  • This flow cell should be ready [DE: i.e. almost shippable]
• Working on recommendation list
• Radically simplifying the software
  • A team is working on redesigning the user experience
  • People should just know how to run MinKNOW
  • After a time, all base calling improvements will be in MinKNOW
  • Process should be very straightforward

VolTRAX

• Moving along pretty well
• After a time all protocols will be moded onto VolTRAX
• By the end of the year, should be using lyophilised reagents
• Multi-sample flow cells
• Quantification + possibly amplification
• Available at the end of the year (VolTRAX mk 1)

Raw signal

• Another slide Clive hates
• Clive doesn’t like the evnets
• Breaks down at fast speed
• Event data bloats the files

Algorithm updates

• 1D will improve
• 2D idea: template + hairpin + complement
  • need to replace with something new
• 1D²; make the second strand hang around for longer so second strand follows through
• 96% modal accuracy, limited by how data is combined
• Should get to 1% error
• 1D² will be implemented next week

R9.5

• New pore
• Backwardly compatible
• Coming 8th May
• All ready to rock
• 2D can now go into the dustbin of history
  • Nanopore killed 2D today (please tweet)
• Very little is different between R9.4 and R9.5
  • What is different should be fixable in software

Homopolymers

• If taken out, now getting pretty good numbers
• Can do quite well, pretty much all homopolymer calls are out by 1
• Homopolymers will go out as systematic errors

Basecalling

• Still an enormous amount of headroom
• End up with detritus of error
• Should get to 99.9% by the end of this year (possibly this summer)
• Still lots of unmodelled signals
  • e.g. there is a fungus that puts sugars on DNA
  • People here have shown that things are visible in the signal

New pores

• Also working on new pores; patent has just been published on this
  • In the future, will now never replace pores, just migrate pores
• Updated pore has two times read-ahead
  • should be able to span 30bp homopolymers
  • may have new pore by the end of summer

Raw Calling

• A lot of work done on this
• Captured on FPGA
• Pretty close now to comparable performance to what is done now
• Making available raw data base caller now
• Any useful software will migrate to MinKNOW
• Want to just output FASTQ

Removing cold-chain system

• Can now store and ship new packaging in wool
  • Creating jobs for sheep
  • A reasonable user will be able to knit their own clothes using nanopore packaging
• A lot of work being done to make a lyophilised reagent system
• Performance is comparable to conventional wet reagents
• 1D² version will be available shortly

Pipettes

• That leaves pipettes; pipettes have to go
• Need to encapsulate prep into a dry / warm container
• Reagents put into prototype tube
• There will come a point when you can put a swab in, and 15-20mins later start looking at reads
• Not quite there to show with date or product yet

Increasing sensitivity

• Can currently do well down to about 1ng
• can go even lower (with reduced yield)

Additional device with new name needed

• Dongle with FPGA and firmware
• Can basecall MinION sequences
• Encapsulated supercomputer
• Computer can be cheap and small
• Device will eventually be able to work on its own (with connected power supply)
• Should be out by the end of summer
• FPGA currently Intel Arria 10, runs at 10Mb per second
  • Next version will be 7Mb per second
• Theme is portability: want to make things as easy as possible

Flongle

• Electronics moved onto dongle
• One-off buy, can be thought of as a mkII MinION
• Cheaper flow cell snaps into adapter
• Fewer channels, higher volume [DE: presumably production volume], cheaper price
• The substrate device… the MinION they should have made
  • Very significant for ONT
  • Should be out by the end of the year
• Previously had gel-gel interface
  • Can now get metal/metal interface
  • have prototype chips and membranes at the other end of the scale
  • was actually a prototype used for the SmidgION

SmidgION

• Won’t be around this year, but it will come
• Chris has been able to run a MinION on a mobile phone

Hypothesis-driven sequencing

• Looking to develop this
• There are a lot of things that can be done with nanopores
• Blob counting: sequencing of site-specific things
  • Let sequences whizz through the pore
  • Blob does the counting
  • Very rapid yes/no sequencing
  • Getting to very rapid MinION sequencing

Solid-state nanopores

• Will be solid-state Flongle; will definitely [DE: also] be a blob counter

Cas9

• Can be programmed: a programmable blob
• Target certain fragments
• Use chemistry to select out molecules for sequencing
• Can use the same trick for detection as well
• Have pretty good working implementation of Cas9 target selection
• Multiplexing
  • Very cost-effective for small region for thousands of samples
  • “Cas Me If You Can”
  • Custom Cas9 provided by ONT
  • Also planning to make pre-made versions

Dates

• May
  • MinKNOW 1D² 9.5
  • 8th: new flow cell
• June 17th (Clive Brown’s birthday): raw basecaller
• June 30th: new cDNA kits
• July: new ambient shipping methods
• September: Cas Me If You Can

Epitome

• Make standard Bioinformatics workflows
• Platform is now pretty mature

Is Basecalling Necessary?

• Do you need to basecall and align?
• Have been teaching neural networks to look at raw signal to decide what species it is
• Want to go from squiggle
• There are cases that don’t require base calling
  • A sequencer that doesn’t do base calling – Clive’s pet project for the rest of the year

Questions

• Gordon’s ambition is to put the Flongle in as a diagnostic device
• Consensus reading errors: just a matter of knocking downm the obvious problems
  • If ONT runs into non-software problems, just need to change the chemistry

Day 2 (May fhe Fifth)

Gordon Sanghera

Throughput

• Since shifting to the 9 series, there has been an exponential move to throughput [DE: it was exponential prior to that, but no one cared because it was exponentially low]
• A lot of issues will be application-specific
• 20 gigabases is the output of a 2-day flow cell, but ONT wants to move to 3 days
• Hope to get all customers to 15-20 gigabases
• No reason why the flow cells can’t reach their theoretical maximum
• Sensor chip is designed to handle 1000 bases per second
  • Might be able to adjust sequencing speed in the future for application-specific uses
• Is there an underlying single molecule optical[?] limit to accuracy?
  • Accuracy doesn’t seem to have a limit with nanopore
  • 100% Accuracy is the target
  • No fundamental reason why this is not achievable
• R9.5 is squeezing the last 10% out of the flow cells

Marketing

• The challenge is getting the technology to market
• In 2017, we might be ready for prime time
  • We are answering biological questions that other platforms cannot solve
• Innovating quickly: it is in our DNA
  • ONT cannot rest on their laurels
  • We are disruptive innovators

Flongle

• This will be big
• Thinking about crossing the chasm into diagnostics [repeated multiple times]
  • If you could, you would probably want to sequence the whole of HIV
• Flongle allows ONT to be applied
  • Can be HIV or HepC
  • Can delete what is not needed
• Will challenge PCR diagnostics, a tech that has been around for 25 years
• Next year, flongle will be available as a registered diagnostic device

Disruption

• The challenge is to disrupt the diagnostic market [etc….]
• DNA information will disrupt everybody
• [video about student education; students having the opportunity to use the MinION]
• Children will say, “You didn’t know that beef was actually Kobi beef, you just ate it?”

Nick Loman

• Setting expectation levels corretly for this talk: The answer is NO! [DE: referring to a statement about the suitability of the MinION in the title]

Dr Seuss

• People with young kids use kids books for inspiration
• The MinION has sequenced almost everywhere
  • On a boat
  • With a goat
  • On a plane
  • On a train
  • Down some holes
  • At the poles
  • In outer space
  • At quite a pace
• One notable place where it hasn’t sequenced; more aboout that later

2015

• Nick’s last talk at London Calling was in May 2015
• Very excited, because they had done successful sequencing in New Guinea, with results back on a time scale of days
• One genome was sequenced per flow cell
  • Generated sequence quickly enough to feed back to WTO
  • By May, had sequenced 50% of Ebola cases in Guinea
• Really important findings came out of that ebola work
  • Data needs to be shared as quickly and as ethically as is possible
  • Could measure cross-border transmissions
  • When a new case popped up, could identify if it was linked to other existing cases

Ebola progression

• It was originally thought that Ebola was over by September / October, but a number of additional flare-ups happened
  • Used MinION to identify one case
  • The seminal fluid of a survivor was infected
  • The virus was “frozen in time”
• [Video / animation: 1600 Ebola Virus transmissions, representing 5% of all outbreaks, 30,000 cases, 10,000 deaths]
• 10.1038/nature22040
• Many opportunities to stop Ebola from moving
• MinION/NickSeq wasn’t deployed until April 2015
• Is there a case for getting sequencing done early?

Zika Project

• Introduced last year
• Unbelievable arrogance in the grant application: claimed that they would sequence 750 genomes
• For epidemiology, see Oli Pybus’ talk

Concentration issues

• Zika was almost impossible to sequence using normal procedures
• Doesn’t work with metagenomic techniques
• cT values for PCR of 34 to 36
• Pathogen enrichment challenges
• At the moment, 100ng of DNA is required for MinION sequencing
• Can be done by PCR, whole-genome amplification, or bait probes
• PCR-based approach represents a pragmatic, or easy way

PCR Primer design

• Josh spent a couple of months working on sequencing
• Developed “Primal Scheme”: an application to generate a multiplex panel for siling sequencing
• Can now sequence Zika, with reasonably complete genomes up to a cT of 36
• Some amplicons drop out more than others
• Now have everything in place to do this outbreak stuff in real time

Yellow Fever

• Currently an outbreak in Brazil
• Process is much quicker this time
  • Biggest hold-up was taking about a week for the university to process the primer order (actually not that bad)
• Josh and Nick didn’t have to go this time
  • Can now enable colleagues to do this
  • Could do six genomes per flow cell, 1-4 million reads
  • Yellow fever has a low cT, so easier to sequence
• Now have easy-to-install workloads

Health Service Publicity

• Check out Nick above the escalators at Euston

Working with high-yield MinION flow cells

• Mix rapid kits together, pull out single contigs
• Yield improvements make this stuff very feasible
• Someone should sequence inside whales
• There is a mapping from whale weight to read length
• Still haven’t got a megabase read
  • A flow cell is running outside, Nick has been checking during the conference
  • Next improvement to be trialled: extracting E. coli from an agarose plug
  • Have got read N50 up to 150kb
  • Can get an entire E. coli genome in seven reads
  • longest read so far is 886kb
• Could we trivially finish a genome?
  • With a 30X dataset, could probably get human contig N50 to 50Mb

Wish list

• All of Nick’s wish list from 2015 is done, except for low-input

NHS – where sequencing is not done

• Sort it out and bring it to the UK
• Money is an issue
• Clinical microbiologists have been reluctant
• Physicians seem quite into it
• A lack of will; people just need to go and do in

Questions / Answers

• Nick hasn’t looked at why people are resistant
• Could do pathogen discovery and RNA at the same time

Lightning Talk – Niranjan Nagarajan

Human Gut microbiome

• Large ongoing project
• Seven questions
  • Host/microbial factorss
  • Evolution of resistome
  • Novel plasmids
  • Strain dynamics
• Begin with a pilot project, using long reads for metagenomic assembly
• Then use real clinical samples
  • The aim is to maintain the diversity of the community
  • Many reads are longer than 5kb
• Carried out an analysis on a single patient
  • Significant enrichment of K. pneumoniae with different abundances
  • N50s of hundreds of kilobases
  • Were able to identify extended-spectrum lactamase gene
• Observe plasmid dynamics and rearrangements in plasmids

Lightning Talk – Michael Boemo

DNA base analogues

• Use analogues to look at DNA replication
• Pulse of thymidine analogues, then can work out where the origins of replication are in the genome
• How well does it work?
  • Analogues make quite a difference in the signal
  • Working pretty well in synthetic substrates
• Don’t care about the exact base, just interested in the region

Lightning Talk – Celine Bigot

Free Pathogen Identification

• NGS enables the detection of microbial species without a-priori models
  • MiSeq, MinION, PGM were all tested
• Working on own reference material, plus the Zymo community standard, plus other types of samples
  • CNRGH standard has 10 bacteria, one fungal, and one other
  • Standard is still in progress
• Interesting preliminary analysis
  • Has done a WIMP analysis
• Based on processing time, the MinION and MiSeq are preferred to the PGM

Lightning Talk – Scott Gigante

Basecalling (Nanonet)

• Custom neural net basecaller; replaced with black box in the slides
• Wants to do away with Hidden Markov Models
• Used two E. coli libraries, one treated with methyltransferase, one unmethylated by PCR
• Used EventAlign for associating the basecall with events
• Training neural networks is not easy
• Feedback used with training, issues are often only realised weeks later
• Using a longer kmer results in a faster training loss, but lower error with increasing kmer size
• It is hard to judge when the training is complete
• GLM calls 99% of sites correctly
• Nanonet provides an excellent opportunity for training

Lightning Talk – Ben Matern

Identification of HLA Variants by cDNA sequencing

• Genes are highly polymorphic
• Interesting to look at the DNA and RNA
  • RNA has alternative splicing
• Used GMap, which can insert introns (splicing-aware mapper)
• Has made a software platform, an alternative to laboratory techniques
• Aligned boundaries mapped to the reference genome
• Can take expression profiles and cluster into expression patterns

Lightning Talk – Benjamin Istace

Banana Genome

• One of the most sold / produced fruit around the workd, 100 million tonnes
• Most edible species of banana are from two hybrids
• Sequenced and assembled two genomes, looking at one now
  • Long fragments were selected with the BluePippin
• Ran four R9.4 runs, generating 22 gigabases of reads
• Peak [modal] read length was between 25-30kb
• Read correction with Canu, processed into SmartDenovo
• Produced a genome with 704 contigs, N50 of 1.85 Mb

Lightning Talk – Matthew McCabe

Sequencing of BRD viruses

• Using the rapid sequencing kit
• Over 20 viruses that are associated with disease
• Diagnosis at the moment is mostly by PCR
• Looking for a rapid, untargeted sequencing approach
• Can untargeted MinION do this?
  • 4 cultures, each infected, pooled, and nuclease added to remove host DNA
  • Non-exised nucleic acid goes into the RAD kit
  • Got 7,000 reads in the pass folder
  • BLAST top hit took 9 hours
  • Local viral sequence took about 10s
  • 99.6% of reads were correctly identified

Lightning Talk – Libby Snell (ONT)

Direct RNA Sequencing

• Kit is available now from the ONT store
• Sequencing of RNA molecule directly; it’s possible for things that maybe couldn’t have been done previously
• Not possible using any other technology
• Up and coming… ONT wants to get rid of cDNA; it’s only there while they’re stabilising the technology
• Ideally want to move to a 30-minute rapid preparation
  • Ligate, wash, and load
• Potentially allows for a lower input amount
• Tested with quantitative samples: Ambion ERCC, Lexogen spike-in (SIRVs)
• Single-stranded prep works, and a 2D prep works (with RNA stabilised by cDNA)
• Getting full-length transcript coverage
  • Currently output is about 70% of what it should be, probably due to blocking

Direct RNA Breakout – John Tyson

[DE: I missed this; I was distracted talking to David Stoddart about chimeric reads]

Direct RNA Breakout – Rachael Workman

Using C. elegans for direct RNA sequencing

• comparing direct RNA to cDNA
• looking at basic differences, what to expect
• Used strand switching and ligation sequencing
• No need for PCR when working with C. elegans, can get 30µg cDNA easily
• Called datasets with both older and newer basecallers

Yield

• Direct RNA had less yield (as expected)
• Direct RNA had shorter reads
• RNA transcripts were perfectly fine for C. elegans
• lower MAQ alignments with RNA
• Once MAQ filtered, match percentage was similar between RNA and cDNA

Comparing to Wormbase transcripts

• cDNA found most transcripts
• 1140 transcripts were detected in RNA, but not in cDNA
• Filtering out by base length didn’t throw out much

Full-length transcripts

• More degredation in the RNA run
• Looking at RNA alignments, get degredation in the UTR
  • Might be aligner clipping
• Other end, more exon dropout, need to look at the soft clips

Abundance comparison

• RNA tracked cDNA fairly well (R=0.76)
• Would be interesting to look at an Illumina run for comparison

Homopolymer calling

• cDNA and RNA were pretty identical
• Initially compared without homopolymer correction
• Also compared with transducer basecalling
• [DE: With the transducer, the amount of C homopolymer bases exactly matched the amount of G bases]

Take-home messages

• Library preps were robust for both cDNA and RNA
• RNA will get better
• You could stop doing Illumina, switch to cDNA and be fine

Direct RNA Breakout – Chris Vollmers

Single-cell RNASeq

• Using nanopore technology to improve
• Human cell atlas is now picking up pace
  • Looking at the types of every human cell
  • Illumina is good at defining gene expression of single cells
• Most single cell library preps involve full-length cDNA as an intermediate step

B-cells

• Lab’s primary interest is in B-cells
• Would love to do direct RNA seq, but cells contain about 10 femtograms of RNA
  • A couple of orders of magnitude less than what is needed
• Was waiting [a long time] for Illumina run results to get back
  • Why not put cDNA into a MinION flow cell?
  • Designed indexes by randomly generating 60bp random sequences
• Trying to quantify gene expression, ended up breaking tools
• Looked at the overlap between Illumina and Nanopore expression
  • Got fairly good correlation, r = 0.9 approx.
  • Nanopore doesn’t seem to have a length bias
  • Cutting is needed for short transcripts, but can’t be done
  • [Possible that the inability to cut small RNA molecules means they are under-represented in Illumina reads]
• Used SIRV / synthetic spike-in molecules as a control
  • Made 10fg of DNA, trivially easy and fairly quantitative
  • Wanted to get by without reference genome annotation

Pipeline – MandalorION

• Only wrote this because it wasn’t available elsewhere
• Picked up stuff [transcript isoforms] that wasn’t in the reference sequence
• Sort reads based on the distribution, create [model] isoforms
• Get approximately linear correlation (r=0.97) when comparing reference-free versus reference-based mapping
  • CD20 gene – found an exon that wasn’t in the reference assembly
  • Looked at assembly by Trinity, found assemblies that were just really bad
• Working on systematic way for naming isoforms [odd that this doesn’t exist already]

Direct RNA Breakout – Andrew Smith

Direct sequencing of 16s rRNA

• Very abundant in cells
• Getting to 16s RNA levels from gDNA would need 13-14 rounds of PCR
• If we could get a hold of 16s, woud be able to do a lot of good things
• Needed to customise the ONT RNA kit a bit
  • Kit adapter was modified to be complementary to the 3’ end of 16s
  • This is a very well-conserved gene at the 3’ end

Proof of principle

• Working with highly-purified 16s
• 1.5kb rRNA gene
• Proof of concept carried out first on E. coli
• For methanococcus sequencing, needed to tweak the adapter a little bit
• With increasing read length, it becomes very much easier to classify things
• Reads that were over 1000 bases had 98% classification accuracy

Coverage hump

• Hump at a particular region
• Very consistent base miscall
• Compared MRE600 E. coli with RsmG-delta
  • Base call was different
• Nanoraw showed a modified ribonucleotide in the 16s gene
  • This is a mechanism by which some microbes can defend against other microbes
• Wanted to show that this could be detected at other regions
  • Pronounced low-amplitude signal from the modified strain
• Also found pseudouridine modification
  • Very exciting to be able to detect this
• Ion current change affects current at adjacent regions
  • Probably due to 5-6bp nucleotide window from which the current is sampled

Diagnostic tool

• Titration with lower and lower amounts of 16s
• Could still see reads even at 5pg
• Depending on the amount added, could find result in 20s
  • Basically the earliest opportunity for a read sequence to appear was when the first 16s reads came through

Direct RNA Breakout – General Discussion and Questions

• Should we stop calling RNASeq [the stuff done on other systems] RNASeq?
  • We are fighting an uphill battle
  • The community will eventually come to a consensus sequence
• Neither GMap nor BLAT use a [transcript] reference; they don’t rely on genome annotations
• Just doing cDNA is losing a lot of information
• Read-until could get rid of ribosomal sequences
• Direct RNA with a polyA primer is the only thing that can get the correct length of the polyA sequence
  • No need to do VT23, can just use T23
  • Can tell mispriming events when they happen
  • [DE: Why not use a polyT primer with a double-stranded component?]
  • SIRVs polyA tail is [always] 30bp
• mRNA only makes up 15% of total RNA sequences
  • There is lots of rRNA
  • Also a whole number of small RNA molecules

Plenary Panel – Raymond Hulzink, KeyGene

Background

• Wet lab biologist
• Started in 2014 with the MinION
• In 2016, started doing serious MinION sequencing
• In 2017, started sequencing the melon genome, and some BAC clones
• Recently obtained the PromethION, have done a lambda trial
• Will use VolTRAX for plant genomic DNA

Plant / crop genomes

• Large, usually polyploid
• Long reads should resolve the genome complexity
  • As read length is increased, contig length increases
• Plant cells are challenging
  • Very rigid cell wall, a bit in contradiction to what needs to be done
  • Have lots of metabolites, can damage DNA or reduce the efficiency of sequencing
  • Have chloroplasts and mitochondria
• DNA isolation has no generic protocol; every plant is different
• Understanding DNA quality is important
  • Can degrade over time
  • Comparison of fresh sample vs. ultra-pure sample, no difference in high-quality reads regardless of the isolation method or storage
  • QC changes when looking at reads longer than 50kb

Melon sequencing

• Decided to use DNA that was less than one month old
• Other crops may need purification
• Compared 75% of the shortest reads to 75% of the longest reads
• Longer reads have contigs over twice the length of the shortest reads

Generic workflow

• Isolation of nuclei
• Want to hunt for ultra-long reads
• Use agarose-embedded nuclei
• Needs a lot of work in the lab to get ultra-pure DNA
• If the fragment length is greater than 25kb, typically end up with read lengths of greater than 15kb
• New albacore improves the read quality
• Expect to generate platinum assemblies from plants

Plenary Panel – Ivo Gut

• This sequencing stuff is a bit of a clown act; Ivo prefers to play guitar
• Doing de-novo sequencing on the hummingbird and houbara bustard

ONT History

• ONT has actually been around since 2008, and collaborated with Ivo Gut at that time
  • Ivo got a ball of dirt as a prize
• Ivo helped in the development of structures to do de-novo assemblies
  • Fosmid-pooled sequencing for a complex genome
  • Used to take a whole lot of complicated preps and bring it all together
  • Required some complicated manoevering and sequencing
  • Has gone on to throw every technology at it
• As time when on, Ivo got more and more confident about nanopore
• At some time, just took DNA and put it into nanopores

Houbara bustard

• Endangered species: lives in the arid desert and is flightless
  • Hunted using other birds (which prefer airplane travel)

Hummingbird

• Part of the genome 10k project, which turned into the vertebrate genome project
• Has been sequenced using everything except nanopores
• Got a lot of help from ONT for sequencing this bird
• Generated about 20X coverage
• Taking only Nanopore reads, got an N50 of 2.7Mb
  • Compared to PacBio, which was 5.37Mb N50 (but about 3 times more data)
• Used a Canu assembly combined with Pilon

Houbara bustard

• Did phenol chloroform extraction
• 10X coverage by nanopore, together with Illumina sequencing
• Got scaffold N50 of 5Mb
• Last night, an additional 3 flow cells were mapped on top
• N50 of 8Mb, with 181 contigs of over 1Mb

The hummingbird now

• Re-exttracted DNA, got tissue shipped to Simon Mayes
• As a pilot, Simon got some duck tissue for sequencing
• Very optimistic that it can be done with long reads
• With the MinION, there’s no more difference between scaffolds and contigs

Plenary Panel – Christiaan Henkel

Genomes that have not been previously sequenced

• Many amphibians and plants with extremely large genomes
• True long-read sequencing
  • If 1Mb is the length of the shard, then the genome is the length of the sun and back
  • Get unique bits in a sea of repetitive content
• Lightweight assembly
  • Assume perfect long reads
  • Just compare the ends, then assemble the genome
  • Doesn’t work for repetitive regions at the end
  • So… find unique regions

Sequencing the eel

• Want to know why they are critically endangered
• Did an assembly: 860Mbp assembly, 2366 scaffolds with 1.2Mbp N50
• Corrected with Pilon
• Tried to compare against an older Illumina-based draft
  • Lots of tiny scaffolds unmapped
  • Some discontinuities, nanopore data almost always compares favourably

Tulip assembly with Tulip

• Library assembly method called Tulip
• Tulip production is easy, but tulip breeding is not
  • Reading a single chromosome would take a month on nanopore [serially-sequenced]
  • It is hoped that the PromethION will be good enough
• Pilot with one MinION flow cell for sequencing
  • Surprisingly, the tulip genome is not very repetitive at all
  • Assembly should work well with the lightweight method

Plenary Panel Questions

• Pilon polishing is a stop-gap measure
• Using other technologies (e.g. 10X, dovetail)… looking at that, but Ivo prefers using as few technologies as possible
• Data analysis – No best practise for de-novo assembly
• A few very smart people out there looking at different things

Kazuharu Arakawa

Starting with the very basics

• needed to do sequencing, so was investigating Nanopore
• Looked at bacterial draft genome
  • Used MinION reads to finish genomes
  • “how low can you go?”
  • In under 1hr, was able to complete a full genome with one 4.4Mb circular chromosome and one 44kb plasmid

Spider silk

• 4 times stronger than steel
• in nature, about 10µm in diameter
• More elastic than nylon
• it’s weird to have both high strength and high elasticity
  • has the highest “toughness” [combination of strength and elasticity] among all fibres
  • Made from proteins, so it’s re-usable
• If spinning a web with 1cm-diameter silk, it would be possible to catch a plane
  • If that same web were made of steel, the plane would crash and break into pieces
  • If that same web were made of rubber, the plane would break through
• Did an experiment comparing spider silk to carbon fibre and polyester
  • Spider silk was able to handle more weight than both of them
  • Carbon fibre broke fairly quickly
  • Polyester stretched too much and broke

Spiber

• Company set up by former institute members
• Taking spider genes and putting them into bacteria
• Creating clothes like the Moon Parker
• Can be made into sheets, gels, and other non-thread things

Silk diversity

• Spiders make a whole lot of different spider silks
• All silk genes are monophylogenetic (from the same original gene)
• With enough data about the genetics and properties, can work out how amino acid changes affect the properties
• Will sequence 1,000 spiders (for funding purposes)
  • Has already done a lot of those [? over 900]
• Spider genomes are quite long
• Don’t really know what components are in the genes
  • One gene has periodic polyalanines
  • Can only find 10 spider silk genes in GenBank

MinION categorisation of genes

• Can’t do PCR because of repeats
• Next time, got a better yield
• Did whole-genome sequencing at 2-3X coverage, got N50 of about 10kb
• The challenge is to find entire components of spider silk
• Protein fractionation yielded 3 bands that were very specific to drag-line silk
  • Did gland-specific transcriptome
  • Looked at top 50 most expressed genes
  • Novel genes were found that corresponded to the gland in the spider
  • But first, they needed to clone the genes
• Tried cloning, sorted out and cloned two genes
• The other two genes were actually different components of the same gene
  • sequenced 2.4kb
  • use MiNION to finish it
  • no conserved N-terminal sequence

Questions

• Only saw repeat elements within spider silk genes

Zoe

• ONT can probably fit another 200 seats in the venue for next year

Gordon Sanghera

• Get back to your labs and get innovating