long read sequencing - lscc lab talk - fri 5 june 2015
TRANSCRIPT
Long read sequencing
Torsten Seemann
VLSCI LSCC Lab Talk - Melbourne, AU - Fri 5 June 2015
The good, the bad, and the really cool.
Why do we need long reads?
Repeats!
Long reads untangle graphs
Completed genomes
Phased haplotypes
Structural variationThe missing heritability - not just SNPs & indels
Long read instruments
Pacific Biosciences RSII
2015 ARC LIEFw/ Tim Stinear
Installed this week.
Passed testing!
Oxford Nanopore MinION MkI
Successor to Mk0
MinION Access Program Round 2
The up & comer!
PacBio It’s already here and it works.
PacBio - the device∷ It’s big!
∷ Three chunks: compute (left): robotics (top): sequencing (bottom)
∷ A cushion of N2 gas
PacBio - technology∷ Polymerase bound to
bottom of ZMW μ-well
∷ Fluorescent nucleotide incorporation measured in real time
∷ 3 hour “movies”
PacBio: read lengths
Needs careful library prep to ensure DNA is
not overly fragmented!
PacBio: error rate
Single read: 86% 30x Consensus: 99.999%
PacBio: main applications
∷ Finished microbial genomes
∷ Full length cDNA (mRNA isoforms)
∷ Extreme GC sequence
∷ HLA / MHC / KIR haplotyping
∷ Base modifications (methylation)
PacBio: bioinformatics
∷ All in GitHub∷ SMRT Portal
: Nice GUI: Cloud ready: Linux backend: Cluster ready
∷ Cmdline too!
Oxford NanoporeThe new kid on the block.
MinION - the device
PromethION - large scale
∷ 48 separate
flow cells
∷ On board ASIC
∷ Runs Python
Nanopore - technology
Nanopore - types of reads“1D reads”
∷ Template 1D﹕ only fwd stran
∷ Complement 1D﹕ only rev strand
“2D reads”
∷ Normal 2D﹕ mostly fwd, some rev
∷ Full 2D﹕ most of fwd & rev﹕ these are high quality
Nanopore - read lengths
Read length is not limited by technology but by library preparation.
Can get >100kbp reads.
Read length
Nanopore - error rate
∷ 5-mer errors∷ Not modelling
base mods yet∷ Basically
where PacBio was a few years ago!
Percent identity (aligned)
MinION - applications
∷ Same as PacBio plus....
∷ Portable sequencing: in the field eg. Josh Quick in Guinea for Ebola: in hospitals - infection control: monitoring - water/food supply, production facilities: at the GP - pathogen test in 10 min from blood prick?: spit in a home device every morning?
MinION - bioinformatics
∷ Event space -vs- base space: MinION MkI - base calling in cloud (Metrichor): MinION MkII - on device?: PromethION - can choose on-device add-on
∷ Mostly 3rd-party tools - lots of activity: poretools, poRe : minoTour, nanoPolish
Disruptive technologyJust another sequencer?
“Run until” Dynamically adjust sequencing yield
“Read until”
∷ Can access events/bases during reading: remember reads are long 40 kbp: examine first 100 bp say: can decide to stop reading and eject molecule!
∷ This is a killer app!: only want pathogens? eject if human DNA: only want exome? eject if not exonic looking: controlled with Python code
VolTRAX - library prep
A new business model
∷ No capital or reagent costs: Instrument will be free: Flow cells will be free: Only pay for what you want to sequence: Min. $20 and ~$1000 for a 100x human genome
∷ But I’ll scam the system!: Flowcell stats sent back to base: Won’t send you new flow cells if they look unused
How will our job change?
Some things never change
∷ Don’t worry!: 50% of our job will always be converting file formats ☺
∷ But things are improving: Pacbio: HDF5: MinION: HDF5 / FAST5
∷ Can convert .h5/.hd5 to .fastq easily
Read alignment
∷ PacBio: BLASR - Basic Local Alignment + Successive Refinement: BWA MEM - bwa mem -x pacbio
∷ MinION: MarginAlign - sum over possible alignments, HMMs: BWA MEM - bwa mem -x ont
∷ Need to modify variant caller parameters
De novo assembly
∷ Pacbio: HGAP, HGAP2, Falcon, Spades, Celera Assembler
∷ MinION: Spades, Celera Assembler, NanoPolish
∷ Lots of convergence: Similar error models (indels): Long reads, lower coverage - back to the future!
Streaming analysis
∷ We are not going to keep all this data
∷ Extract info we need and discard
∷ Cheaper to resequence?
∷ Need to think streaming analyses
∷ Lots of new applications
Conclusion
Exciting times!
∷ Genomics is changing all the time: new technologies: changing attributes/properties of current technology
∷ Bioinformaticians need to be able to adapt: focus on key skills not specific apps
∷ Pipelines are often short lived: except maybe clinical / accredited ones