ECCB 2010 Next-gen sequencing Tutorial

Tutorial 1: Working with next-generation sequencing data - A short
primer on QC, alignment, and variation analysis of next-generation
sequencing data

Presenters: Thomas Keane and Jan Aerts

Thomas Keane 9th European Conference on Computational Biology 26th September, 2010

Schedule

Time Presenter Topics

9:00-10:30 Thomas •  Overview of next-generation sequencing technologies
•  Applications of next-generation sequencing
•  Quality Control
•  Next-gen data formats
10:30-11:00 BREAK

11:00-12:30 Thomas •  Short read alignment algorithms and tools
•  Practical examples on use of short read aligners
•  Sequence assembly methods and tools
•  Case study: 1000 genomes project
•  Experimental Design
12:30-13:30 LUNCH

13:30-15:00 Jan •  Overview of variation calling from next-generation
sequence data
•  SNP and indel calling theory and tools
•  Practical examples of variation calling
15:00-15:30 BREAK

15:30-17:00 Jan •  Introduction to structural variation
•  Summary of different types of structural variants
•  Algorithms and tools for calling structural variants
•  Visualisation of structural variants

About Me

Joined the Sanger Institute in 2006
2006-2008 Pathogen Genomics Group
  Genome assembly
  Capillary based eukaryote pathogen projects: Malaria, Trypanosoma brucei, Babesia
bigemina
  Assessment of early next-gen data
  25bp illumina reads – determining error rates, concordance with capillary data,
sequence biases
  454 assembly assessment of bacterial and parasitic eukaryote genomes
  Correcting 454 assemblies with Illumina data
2008-date Vertebrate Resequencing Informatics
  Established group with Jim Stalker
  Initial projects
  1000 Genomes project (http://www.1000genomes.org)
  Data releases, alignments, Sanger 1000G sequencing, data submission, aligner evaluation
  Mouse Genomes Project (http://www.sanger.ac.uk/mousegenomes)
  Sanger project, catalogue variants across 17 mouse strains, de novo assembly, data release
and visualisation
  Upcoming projects
  UK10K project
  4000 whole genomes at 6x, 6000 human exomes
  Medical sequencing - Phenotype based study
Email: thomas.keane@sanger.ac.uk


Tutorial 1: Overview, Applications, QC and Formats

 Overview
 Quality Control
 Next-gen Data Formats

 Short Read Alignment

 Sequence Assembly
 Case Study: 1000 Genomes

 Experimental Design

Some Perspective

MR Stratton et al. Nature 458, 719-724 (2009)


Previous Technology: Sanger Sequencing
1992-1999
ABI 373 / 377 Slab gel sequencers
2-3 runs per day, 36 - 96 samples per run
100kb per instrument per day 80 staff
to operate

2000
ABI 3700 Capillary sequencer 8 runs per
day, 96 samples per run
400kb per instrument per day 40 staff
to operate

2004
ABI 3730xl Capillary sequencers
15 - 40 runs per day x 96 samples
1-2Mb per instrument per day
4 staff to operate

A view from the Sanger

  2000-2003 peak monthly capillary:
~2.7Gb
  August ‘08 production: ~278Gb
  August ‘09 production: ~1.38Tb
  Jan ‘10 Illumina
 HighSeq 2000 – min 200Gb per run
 ~60x human depth per run


A View from Sanger


Next-gen Sequencing Technologies

Turner et al., 2009

The Sequencing Iceberg

20-60Gb results data
per run

1Tb of images, One sequencing
intermediate run every 3 days
analysis files and 56Tb of raw data (per instrument)
tracking data per to collect and
run process each 28 instruments
week!


Storage And Compute

450Tb “staging” area to
collect raw data
800 CPU compute farm
  primary data processing
  Secondary analysis
  Alignment
What do we store?
  SRF files for
submission
 8-10 bytes per bp Need real picture!!
  Fastq file of reads
 1 byte per bp
  BAM files of reads
 1.7 bytes per bp
 Aligned and unaligned
reads
 Second best base call
 Original + recalibrated
base qualities


Next-gen Applications

Whole Genome Shotgun Sequencing (WGS)
  Randomly shear whole DNA and sequence fragments
  Paired end sequencing
Targeted/exome sequencing
  Randomly shear whole DNA
  Select fragments based on some pre-defined set of templates (“baits”)
  Exome sequencing targets the coding regions
  Requires pre-design of pulldown assay
  Currently human, mouse, zebrafish
RNA-seq
  Start with whole RNA
  Reverse transcribe to cDNA, randomly shear and sequence
  Alignment depth is proportional to the relative abundance of the transcript
  Gene finding + alternative splicing
  No pre-design required
ChiP-seq
  ChIP is a powerful method to selectively enrich for DNA sequences bound by a
particular protein in living cells
  Align + peak calling to detect active binding sites
  ChIP-Seq data can be used to locate the binding site within few tens of base pairs of
the actual protein binding site


Large-scale sequencing projects

Capillary days
  Single individual, single genome projects
  Small amounts of variation based sequencing

Next-generation era
  Multiple individuals from multiple populations
  1000 Genomes Project
  MalariaGEN - sequencing thousands of malaria isolates
  ICGC – International Cancer Genome Consortium
  Mouse Genomes Project – sequencing 17 common laboratory mouse
strains
  1001 Genomes Project - Arabidopsis whole-genome sequence variation
  UK10K – sequencing of 10,000 healthy and disease affected individuals


The Next-Generation Process

Randomly shear DNA + end repair + size select
DNA Prep

Append sequencing adapters
Library Prep

Layout of library on sequencing slide or wells
Chip Prep
For each library fragment – determine the order
and identity of bases at either end of the
Sequencing fragment

Raw Analysis Image processing + base calling
  Base calls + associated quality: Fastq/BAM


Sequence Quality Control

Sequence quality control is essential
Why?
  Sequencing runs cost money
 Sequencing a poor library on multiple runs – throwing money away!
  Cost of analysing data
 CPU time required for alignment and assembly
 Cost of storing raw sequence data
 People time in analysing interpreting the results from poor data
Two aspects of sequencing QC
  Run/Lane QC
 Is the data off this lane useful?
  Library QC
 Is this library worth sequencing more lanes off?
 E.g. Have 2x but want another 8x in total
  Slightly different set of metrics


1000 Genomes Pilot Data

Gb submitted

1600

1400

1200

1000
Gb

800

600

400

200

0
454MSC ABHTD BCM BGI BI ILLUMINA MPIMG SC WUGSC
Sequencing Centre


1000 Genomes Pilot Data

Gb submitted Gb mapped

1600

1400

1200

1000
Gb

800

600

400

200

0
454MSC ABHTD BCM BGI BI ILLUMINA MPIMG SC WUGSC
Sequencing Centre


Raw Sequence QC

DNA sequencing is not an exact science!
Quality control on all sequence data/libraries is essential
1 test lane/plex per sequencing library
  Pass QC then sequence more from the library
Run our own QC pipeline within our group
  GC plot vs. reference GC
  Adapter count
  Sample 100Mbp from the lane
 Align to reference
 Percent reads/bases mapping
 Error rate
 Insert size plot
 GC vs. depth plot
 Duplication rate
 Clipping points


Lane QC


GC


Average Base Quality


Insert size


GC-Depth


Genotype checking

How do you know the data you get out is from the individual you
meant to sequence? Mistakes happen:
  Original sample might be wrong individual
  Library might be made from wrong sample
  Lane might be loaded from wrong library

Check:
  Compare to known genotypes
 samtools pileup + glf checkGenotype => likelihood that lane belongs to all
samples
 3-4% of lanes found to be incorrect
  Compare to other lanes of same library
  Barcode libraries
 Yield from single lane now ~1-1.5x
 Allows for multiplexing of samples within a lane
  Up to 24 libraries per lane with barcodes


Library Duplicates

All second-gen sequencing platforms are NOT single molecule
sequencing
  PCR amplification step in library preparation
  Can result in duplicate DNA fragments in the final library prep.
  PCR-free protocols do exist – require large volumes of input DNA
Generally low number of duplicates in good libraries (<3%)
  Align reads to the reference genome
  Identify read-pairs where the outer ends map to the same position on the
genome and remove all but 1 copy
 Samtools: samtools rmdup or samtools rmdupse
 Picard/GATK: MarkDuplicates
Can result in false SNP calls
  Duplicates manifest themselves as high read depth support


Library Duplicates


Duplicates and False SNPs

NA12005 - chr20:8660-8790

Activity 1: Sequence QC

There are a set of human 1000 genomes lanes with QC information
for each lane

Which lanes should pass/fail QC? And why?

Pass criteria
  >80% of reads mapping
  <5% duplicates
  <2% error rate
  Single fragment size peak
 Automated measure: 80% of reads within 100bp of peak
  Single GC peak


Lane 1


Lane 2


Lane 3


Lane 4


Library QC


Real Example from 2009

Evidence
  Many sequencing lanes showing increasing GC levels
  GC-depth plot showing skew towards higher GC regions of the genome
  Resulting in more false positives for structural variation calling
Possible sources
  Library preparation
  Type of library prep. method
  Library prep. materials changed
  DNA shearing method
  Sequencing machine or chemistry issue
  Sample
  DNA poor quality/biased
Key questions
  Are all library types affected?
  Are all runs affected?
  Are all DNA samples affected?
Final Proof
  An individual sequencing library sequenced in January, March, July showed increasing
levels of GC in the lanes off the individual library
Explanation: Vendor had changed the chemistry on the machine that resulted in
GC richer clusters on the slide


Base Quality Recalibration

Each base call has an associated base call quality
  What is the chance that the base call is incorrect?
 Illumina evidence: intensity values + cycle
  Phred values (log scale)
 Q10 = 1 in 10 chance of base call incorrect
 Q20 = 1 in 100 chance of base call incorrect
  Accurate base qualities essential measure in variant calling
Rule of thumb: Anything less than Q20 is not useful data
Typically phred values max. out at Q35-40
Illumina sequencing
  Control lane used to generate a quality calibration table
  If no control lane – then use pre-computed calibration tables
Quality recalibration
  1000 genomes project sequencing carried out on multiple platforms at
multiple different sequencing centres
  Are the quality values comparable across centres/platforms given they have
all been calibrated using different methods?


Base Quality Recalibration

Original recalibration algorithm
  Align subsample of reads from a lane to human reference
  Exclude all known dbSNP sites
 Assume all other mismatches are sequencing errors
  Compute a new calibration table bases on mismatch rates per position on the
read
Pre-calibration sequence reports Q25 base calls
  After alignment - it may be that these bases actually mismatch the reference
at a 1 in 100 rate, so are actually Q20
Recent improvements – GATK package
  Reported/original quality score
  The position within the read
  The preceding and current nucleotide (sequencing chemistry effect) observed
by the sequencing machine
  Probability of mismatching the reference genome
NOTE: requires a reference genome and good catalog of variable sites to
exclude these sites


Base Quality Recalibration Effects


Base Quality Recalibration Effects

Potential Explanations
  an excess of mismatches and often at the beginning and end of the reads
  Illumina is now using a training procedure which treats the first five cycles in a read differently – some
of this affect is real
  adapter sequence being seen at the ends of reads – high mismatches vs. human reference
Latest GATK modified to take these into account
  http://www.broadinstitute.org/gsa/wiki/index.php/Base_quality_score_recalibration



 Overview





Next-gen Data Formats

Raw Data
 Initially stored raw images (tiff files)
 Millions of images generated per sequencing run
 Analyse and delete!
 SRF
 Submission to archives (ERA/SRA)
 Wrapper for the technology sequencing specific output files
 Fastq
 Read sequences and base qualities
 SAM/BAM
 Read sequences, qualities, and meta data
 Multiple technologies, libraries, lanes
 VCF
 Variant call Format
 Store SNPs, short indels, and recently extended for structural
variants


Fastq

FASTQ has emerged as a de facto file format for sharing sequencing
read data
  Simple extension to the FASTA format:
  Sequence and an associated per base quality score
Original Sanger standard for capillary data
Format
  Subset of the ASCII printable characters
  ASCII 33–126 inclusive with a simple offset mapping
  perl -w -e "print ( unpack( 'C', '%' ) - 33 );”


SAM/BAM Format

Proliferation of alignment formats over the years: Cigar, psl, gff, xml etc.
SAM (Sequence Alignment/Map) format
  Single unified format for storing read alignments to a reference genome
BAM (Binary Alignment/Map) format
  Binary equivalent of SAM
  Developed for fast processing/indexing
Advantages
  Can store alignments from most aligners
  Supports multiple sequencing technologies
  Supports indexing for quick retrieval/viewing
  Compact size (e.g. 112Gbp Illumina = 116Gbytes disk space)
  Reads can be grouped into logical groups e.g. lanes, libraries,
individuals/genotypes
  Supports second best base call/quality for hard to call bases
Possibility of storing raw sequencing data in BAM as replacement to SRF &
fastq


Read Entries in SAM

No. Name Description
1 QNAME Query NAME of the read or the read pair
2 FLAG Bitwise FLAG (pairing, strand, mate strand, etc.)
3 RNAME Reference sequence NAME
4 POS 1-Based leftmost POSition of clipped alignment
5 MAPQ MAPping Quality (Phred-scaled)
6 CIGAR Extended CIGAR string (operations: MIDNSHP)
7 MRNM Mate Reference NaMe (‘=’ if same as RNAME)
8 MPOS 1-Based leftmost Mate POSition
9 ISIZE Inferred Insert SIZE
10 SEQ Query SEQuence on the same strand as the reference
11 QUAL Query QUALity (ASCII-33=Phred base quality)

Heng Li , Bob Handsaker , Alec Wysoker , Tim Fennell , Jue Ruan , Nils Homer , Gabor Marth , Goncalo Abecasis ,
Richard Durbin , and 1000 Genome Project Data Processing Subgroup (2009) The Sequence Alignment/Map
format and SAMtools, Bioinformatics, 25:2078-2079


Extended Cigar Format

Cigar has been traditionally used as a compact way to represent a
sequence alignment
Operations include
  M - match or mismatch
  I - insertion
  D - deletion
SAM extends these to include
  S - soft clip
  H - hard clip
  N - skipped bases
  P – padding
E.g. Read: ACGCA-TGCAGTtagacgt

Ref:
ACTCAGTG—-GT

Cigar: 5M1D2M2I2M7S


What is the cigar line?

E.g. Read: ACGCA-TGCAGTtagacgt

Ref:
ACTCAGTG—-GT

Cigar: 5M1D2M2I2M7S

E.g. Read: tgtcgtcACGCATG---CAGTtagacgt

Ref:
ACGCATGCGGCAGT

Cigar:


Read Group Tag

Each lane (or equivalent unit) has a unique read group (RG) tag
1000 Genomes
  Meta information derived from DCC
RG tags
  ID: SRR/ERR number
  PL: Sequencing platform
  PU: Run name
  LB: Library name
  PI: Insert fragment size
  SM: Individual
  CN: Sequencing center


Activity 2: Interpreting SAM/BAM files

From reading page 4 of the SAM specification, look at the following
line from the header of the BAM file:

@RG ID:ERR001711 PL:ILLUMINA LB:g1k-sc-NA12878-CEU-1 PI:200 DS:SRP000032
SM:NA12878 CN:SC

What does RG stand for?

What is the sequencing platform?

What is the sequencing centre?

What is the lane accession number?

What is the expected fragment insert size?


1000 Genomes BAM File

samtools view –H my.bam

How is the BAM file sorted?
How many different sequencing centres contributed lanes to this BAM file?
What is the alignment tool used to create this BAM file?
How many different sequencing libraries are there in this BAM? Hint: RG tag


SAM/BAM Tools

Well defined specification for SAM/BAM
Several tools and programming APIs for interacting with SAM/BAM files
  Samtools - Sanger/C (http://samtools.sourceforge.net)
 Convert SAM <-> BAM
 Sort, index, BAM files
 Flagstat – summary of the mapping flags
 Merge multiple BAM files
 Rmdup – remove PCR duplicates from the library preparation
  Picard - Broad Institute/Java (http://picard.sourceforge.net)
 MarkDuplicates, CollectAlignmentSummaryMetrics,
CreateSequenceDictionary, SamToFastq, MeanQualityByCycle,
FixMateInformation…….
  Bio-SamTool – Perl (http://search.cpan.org/~lds/Bio-SamTools/)
  Pysam – Python (http://code.google.com/p/pysam/)


BAM Visualisation

http://www.sanger.ac.uk/mousegenomes

BAM Visualisation - IGV


BAM Visualisation Tools

Requirements
  BAM files over http/ftp, Annotation data, multiple tracks
  Data on demand – only open a window of data at a time
LookSeq
  http://www.sanger.ac.uk/resources/software/lookseq/
  Perl/CGI/Samtools
  Web based
  Connects to BAM files over ftp
  Large genomes
IGV
  http://www.broadinstitute.org/igv/
  Java/Picard
  Standalone application
  Multiple tracks/BAM files simultaneously
  Annotation data via DAS
  Large genomes
Savant
IGB


Recommended Reading

Cock et al (2009) The Sanger FASTQ file format for sequences with
quality scores, and the Solexa/Illumina FASTQ variants, Nucleic Acids
Research, 38 (6)1767-1771
Li et al. (2009) The Sequence Alignment/Map format and SAMtools,
Bioinformatics, 25 (16):2078-2079
McKenna et al (2010) The Genome Analysis Toolkit: A MapReduce
framework for analyzing next-generation DNA sequencing data,
Genome Research, 20:1297-1303
Manske et al (2010) LookSeq: A browser-based viewer for deep
sequencing data, Genome Research, 19:2125-2132
Savant: Genome Browser for High Throughput Sequencing Data
Marc Fiume; Vanessa Williams Andrew Brook; Michael Brudno
Bioinformatics; 2010 Aug 15;26(16):1938-44.



 Overview






Short Read Alignment

Next-gen considerations
  Amount of data many orders of magnitude higher => memory + speed
  Different error profiles than the previous-generation technology
 Roche 454 reads to have insertion or deletion errors at homopolymers
 Illumina low quality calls can occur anywhere in a read
 Increasing likelihood of sequence errors toward the end of the read for ABI
SOLiD + illumina
 Output of the SOLiD machine is a series of colors representing two
nucleotides
Alignment itself is the process of determining the most likely source
within the genome sequence for the observed DNA sequencing read
DNA based alignment mostly
  Very small evolutionary distances (human-human, strains of the
reference genome)
  Assumptions about the number of expected mismatches
 Allow for much faster processing


Short Read Alignment Algorithms

Large and ever growing number of implementations for short-read
sequence alignment
Fundamental algorithms can be divided into two approaches
  Hash table based implementations
  Burrows-Wheeler transform (BWT)
Both approaches apply to sequence and colour space and all
technologies
Heuristics to find potential locations on the genome
  Slower more accurate alignment run on a subset of potential locations
  Similar strategy to traditional read alignment algorithms: Blat, Blast,
SSAHA2
Constant trade-off: speed vs. sensitivity
Guaranteed high accuracy will generally take longer


Hash Table Based Alignment

Hash table is a common data structure that is able to index complex and
non-sequential data in a way that facilitates rapid searching

Hash of the reads: MAQ, ELAND, ZOOM and SHRiMP
  Smaller but more variable memory requirements
Hash the reference: SOAP, BFAST and MOSAIK
  Advantage: constant memory cost


Hash Table Based Alignment

Hash is typically built from a set of seeds spaced along the reference or
read sequence
Seeds used in the hash table creation and the reads have been
associated with the region of the genome
Specialized and accurate alignment algorithm is used to determine the
exact placement
  Gapped and ungapped versions of Smith-Waterman
  Use the base qualities in order to determine the most likely alignment of a
read
  Use information about where the mate pair aligns to
 Maq: Require 1 mate to align ungapped and can do full Smith-Waterman to align
its mate (e.g. indel in the mate)
Mapping quality
  A probability measure of the alignment being incorrect
  Single end mapping quality vs. paired end mapping quality
  Low complexity regions typically have low mapping qualities
  Typically represented as a phred score (i.e. Q10 = 1 in 10 incorrect, Q20 = 1
in 100 incorrect)


Mapping Qualities

LookSeq
http://www.sanger.ac.uk/mousegenomes viewer


BWT Based Alignment

New generation of short read aligners based on BWT
  BWA, SOAP2, BOWTIE
BWT commonly used in data compression – compressed suffix array
FM index data structure
  Ferragina and Manzini - suffix array is much more efficient if it is created
from the BWT sequence
  FM index retains the suffix array's ability for rapid subsequence search
  Similar or smaller in size than input genome
Two step creation process
  Sequence order of the reference genome is modified using the BWT
  Final index is created; it is then used for rapid read placement on the
genome


BWT Based Alignment

Li and Durbin (2009) Bioinformatics for further details

Which approach?

BWT implementations are much faster than their hash-based
counterparts
  Several times faster still at slightly reduced sensitivity levels
  BOWTIE's reported 30-fold speed increase over hash-based MAQ with
small loss in sensitivity
  Limitations to BWT approaches: BWA is only able to find alignments
within a certain 'edit distance’
 100-bp reads, BWA allows 5 'edits
 Important to quality clip reads (-q in BWA)
  Non-A/C/G/T bases on reads are simply treated as mismatches
Hash based approaches are more suitable for divergent alignments
  Rule of thumb: <2% divergence -> BWT
 E.g. human alignments
  >2% divergence -> hash based approach
 E.g. wild mouse strains alignments


Aligner Comparisons

0.09% SNP mutation rate, 0.01%
indel mutation
BWA: Li and Durbin (2009)
Bioinformatics
Varying mutation rates
Stampy: Lunter and Goodson,
unpublished


Alignment Limitations

Read Length and complexity of the genome
  Very short reads difficult to align confidently to the genome
  Low complexity genomes present difficulties
 Malaria is 80% AT rich – lots of low complexity AT stretches
Alignment around indels
  Next-gen alignments tend to accumulate false SNPs near true indel
positions due to misalignment
  Smith-Waterman scoring schemes generally penalise a SNP less than a
gap open
  New tools being developed to do a second pass on a BAM and locally
realign the reads around indels and ‘correct’ the read alignments
High density SNP regions
  Seed and extend based aligners can have an upper limit on the number
of consecutive SNPs in seed region of read (e.g. Maq – max of 2
mismatches in first 28bp of read)
  BWT based aligners work best at low divergence


Read Length


Example Indel


Using BWA

bwa index [-a bwtsw|div|is] [-c] <in.fasta>
  -a STR BWT construction algorithm: bwtsw or is
  bwtsw for human size genome, is for smaller genomes
bwa aln [options] <prefix> <in.fq>
  Align each single end fastq file individually
  Various options to change the alignment parameters/scoring matrix/seed
length
bwa sampe [options] <prefix> <in1.sai> <in2.sai>
<in1.fq> <in2.fq>
  sai files produced by aln step
  Produces SAM output
bwa samse [-n max_occ] <prefix> <in.sai> <in.fq>
  Unpaired reads – produces SAM output
Use samtools to manipulate SAM file or convert to BAM
  samtools view –bS in.sam > out.bam


Using Smalt

Hash based aligner
  Range of sequencing platforms (Illumina/454/Capillary)

Ponstingl and Ning


Using Smalt

Index the genome
  smalt index -k 13 -s 6 hs37k13s6 NCBI37.fasta
  Produces an kmer index of the genome
  -k 13 speciﬁes the length, -s 6 the spacing of the hashed words
Align the reads
  smalt map -i 800 hs37k13s6 mate_1.fastq mate_2.fastq


Parallelising Short Read Alignment

Simple parallelism by splitting data
  Split lane into 1Gbp chunks and align on different processors
 BWA ~8 hours per 1Gbp chunk
  Merge chunk BAM files back into single lane BAM
 Samtools merge command
  Merge lane BAM files for a library into single library level BAM
 Run duplicate removal step (either samtools or MarkDuplicates in Picard)
 Again check the BAM is not truncated!
  Merge sample libraries together into a single sample BAM
 Run variant calling off sample level BAM
Computational issues to note
  Most aligners produce BAM files
  Always check for truncated BAM files
 End of file marker
 Or use samtools flagstat to count reads post alignment


Reading and Links

Flicek and Birney (2009) Sense from sequence reads: methods for
alignment and assembly, Nature Methods, 6, S6 - S12

Samtools: http://samtools.sourceforge.net/
BWA: http://bio-bwa.sourceforge.net/bwa.shtml
Smalt: http://www.sanger.ac.uk/resources/software/smalt/

Local realignment
  http://www.broadinstitute.org/gsa/wiki/index.php/
Local_realignment_around_indels



 Overview






Sequence Assembly

A fundamental goal of DNA sequencing has been to generate large,
continuous regions of DNA sequence
Whole-genome shotgun proven to be the most cost-effective and
least labour intensive method of sequencing
  Human genome completed by a BAC-by-BAC strategy
Capillary sequencing reads ~600-800bp in length
  Overlap based assembly algorithms (phrap, phusion, arachne, pcap…)
  Compute all overlaps of reads and then resolve the overlaps to generate
the assembly
Volume + read length of data from next-gen sequencing machines
meant that the read-centric overlap approaches were not feasible
  1980’s Pevzner et al. introduced an alternative assembly framework
based on de Bruijn graph
  Based on a idea of a graph with fixed-length subsequences (k-mers)
  Key is that not storing read sequences – just k-mer abundance
information in a graph structure


Sequence Assembly

Green, E.D. (2001) NRG 2, 573-583


De Bruijn Graph Construction 1

Genome is sampled with random sequencing 7bp reads (e.g. Illumina or 454)
Note errors in the reads are represented in red
Flicek & Birney (2009) Nat Meth, 6



The k-mers in the reads (4-mers in this example) are collected into nodes
and the coverage at each node is recorded (numbers at nodes)
Features
  continuous linear stretches within the graph
  Sequencing errors are low frequency tips in the graph



Graph is simplified to combine nodes that are associated with the continuous linear
stretches into single, larger nodes of various k-mer sizes
Error correction removes the tips and bubbles that result from sequencing errors
Final graph structure that accurately and completely describes in the original genome
sequence


Next-gen Assemblers

First de Bruijn based assembler was Newbler
  Adapted it to handle main source of error in 454 data – indels in
homopolymer tracts
Many de Bruijn assemblers subsequently developed
  SHARCGS, VCAKE, VELVET, EULER-SR, EDENA, ABySS and
ALLPATHS
  Most can use mate-pair information
Few next-gen assemblers capable of assembling mammalian sized
genomes out of the box
  ABySS – distributes de Bruijn graph over a network of computers using
MPI protocol
  SOAP (BGI) and Cortex (unpublished)
 Key: early removal of spurious errors from the data


Reads Pre-cleaning

Illumina sequencing error rate 1-2% depending on read length
many of the 25-mers will contain errors
Error correction before assembly for small data sets is less important
  Can be removed during the graph assembly
Large datasets
  Removal of singleton kmers is essential as will drastically reduce the
memory footprint of the graph
  e.g. Asian human genome data, the total number of distinct 25-mers was
reduced from 14.6 billion to 5.0 billion

Li et al (2009) Gen Res, 20


SOAP denovo Workflow

Li R et al. Genome Res. 2010;20:265-272


MouseGenomes Assembly – A low memory window
based approach


Gap Filling

Assembly post processing step to close gaps between contigs
Gaps can be created by the assembler where there are changes in the read depth
  E.g. assembler might require a miniumum read depth at node in de Bruijn graph

Tsai et al. (2010) Gen Biol


Gap Filling Algorithm

Non-reference based
  De novo contig extension

Reference guided

Reference


Assembly Evaluation – N50

N50 has traditionally been used to compare assemblies
If you order the set of contigs produced by the assembler by size
  N50 is the size of the contig such that 50% of the total bases are in
contigs of equal or greater size
E.g.
15kb
15kb
12kb
6kb
12kb 9kb 9kb

2kb 7kb
7kb 6kb
5kb 5kb
2kb


Assembly Evaluation – N50

What is the N50?
25kb
19kb
15kb
6kb
21kb
12kb 9kb
2kb
7kb
5kb
4kb
3kb

1kb


Assembly Length vs. N50

Another informative measure is total length of the assembly
  Most genomes have an expected size prior to running assembly
  Assemblers assume diploid genome
Contig total length less than scaffold total length
  Scaffolds are contigs with runs of N’s between the contigs
If you remove smaller contigs -> N50 increases :0)
  Total length decreases i.e. less of the genome sequence in the
assembly :0(
Most assemblers will remove contigs less than 100bp or less than the
read length


Assembly Evaluation Metrics

N50 just measures the continuity of the assembly
  Larger values are generally better
However it does not assess the quality of the assembled sequence
  E.g. if there are incorrect joins in the assembly the N50 could appear to
be larger
Assembly quality measures
  Methods using contigs only:
 N50
 Total contig length
 Number of contigs
  Metrics using an alignment of reads onto the contigs
 Mapping Fraction (No. reads mapped/total reads) + pairing rate
 Count the SNPs and indels
 Misassemblies (regions not spanned by read pairs)
Read pairs

Contig


Which human assembly is better? Why?

Assembly 1 Assembly 2 Assembly 1 Assembly 2
N50 51kb 42Kb 50Kb 20Kb
Total length 2.7Gb 2.69Gb 1.2Gb 2.7Gb
Avg. length 45Kb 39kb 40Kb 18Kb
Mapping rate 0.82 0.78 0.6 0.85
SNP rate 0.02 0.02 0.02 0.02
Indel rate 0.01 0.01 0.01 0.012
Pairing rate 0.8 0.9 0.9 0.88
Misassemblies 15 5 2 2


Which 1Mbp assembly is better? Why?


Recommended Reading

Li et al (2010) De novo assembly of human genomes with massively
parallel short read sequencing, Genome Research, 20: 265-272

Velvet: algorithms for de novo short read assembly using de Bruijn
graphs. D.R. Zerbino and E. Birney. Genome Research 18:821-829

Michael C Schatz, Arthur L Delcher and Steven L. Salzberg (2010)
Assembly of large genomes using second-generation sequencing,
20(9):1165-73

Isheng J Tsai, Thomas D Otto and Matthew Berriman (2010)
Improving draft assemblies by iterative mapping and assembly of
short reads to eliminate gaps, Genome Biology, 11:R41


Case Study: 1000 Genomes Project Workflow

“Sequencing the genomes of at least a thousand people from around the world to
create the most detailed and medically useful picture to date of human genetic
variation”
International multi-institute project
  Sanger, Broad, WashU, MPIMG, BGI, BCM, Illumina, Roche, AB
Data co-ordination centre (DCC)
  Joint NCBI/EBI with the sequence archives (SRA/ERA)

1000 genomes pilot
  Pilot 1: 179 individuals (>2x per individual)
  4 populations
  Pilot 2: 2 high depth trios (40-60x per individual)
  Pilot 3: 697 individuals via exon pulldown sequencing
  7 populations
~4.8Tb data from ~12,000 lanes in ~26,000 fastq files

Main project
  ~1200 individuals at ~4x coverage
  25 distinct populations


Dealing with the data….the early days of the 1000
genomes project

Timothy K. Stanton


Main project: ~1200 samples at 4x in 2009/10

CEU
FIN
GBR
AJM TSI CHB
JPT
IBS
PUR CDX
MXL
YRI CHS
ASW
GWD
ACB LWK KHV
CLM GHN
MAB
PEL

Major population groups comprised of subpopulations of ~100 each

August 2010 Release

No. Samples
120

100

80

60

40 No. Samples

20 3E+12

0
2.5E+12

2E+12

1.5E+12 Total Bp
Mapped Bp
Total Samples: 627 1E+12

Total Bases: 15.6Tbp 5E+11

0
ASW CEU CHB CHS FIN GBR JPT LWK MXL PUR TSI YRI


1000 Genomes Pipeline
(Seq. Centers)
Data submissions (ERA/SRA)

Raw Sequence Data

Reject Data Data quality checking
Genotype checking (DCC)

Fastq, index file

Mapping, Merging & Recalibration (Sanger/TGEN)

BAM files

SNP calls FTP site (DCC)

Analysis subgroups


Sanger Pipeline


Vertebrate Resequencing Informatics Storage

1.5Pbytes of total storage
3 Tiers of storage
  High-performance disk (33%)
 BAM releases, current variant calls (VCF)
  Medium performance nfs disk (66%)
 Lane BAMs
  Low performance backup disk (33%)
 Backup of individual BAMs, old variant call releases (VCF)
M M M
O  O  O  O  O  O 
3G/sec Node n 
Lustre  Lustre  Lustre  Node n 
Node n 
n Tb  n Tb  n Tb  P  P  Node n 
P 
P  P  Node n 
P 
P  P  Node n P 
P  P  P 
P  P  P 
300M/sec
D  P 
D  P  P 
Nfs 01  Nfs 02  Nfs 03  Nfs 01 
D  D 
D  D 
D  D 
D  D 
D  D 
Warehouse (backup) 


Vertebrate Resequencing Tracking


Experimental Design

Choosing right sequencing technology to get optimal results for
experiment

Experiment 1: “I want to determine the genome of a new fungi species
with no closely related reference genome”
  Whole-genome sequencing
  De novo assembly with no reference
 Longer reads might be more useful – 454?
  Mixture of fragment sizes
 200, 500, 3kb, 5kb, 10kb
 Short range pairing information and long range information for scaffolding

Experiment 2: “I want to measure the relative expression level differences
of one yeast species under different environmental conditions”
  Sequence the transcriptome (RNA-seq)
  Illumina or SOLiD sequencing for high depth
 Multiplex the sequencing into a singlle lane
  Measure the relative expression levels by aligning
 e.g. use Cufflinks to detect differential expression across the samples


Experimental Design

Experiment 4: “I want to catalog all of the structual variants in a human cancer cell
vs. the normal cell for as little cost as possible”
  Fragment coverage vs. sequence coverage
  SVs are called from discordant read pairs – long range information
  Sequence coverage not important
  Require fragment coverage
  Sequence multiple paired libraries with short read length
  E.g. 1000bp in total capacity
  100bp reads = 5RPs = 5 fragments x 500bp per fragment = 2.5Kbp fragment coverage
  40bp reads = 25RPs = 25 fragments x 500bp per fragment = 12.5Kbp fragment
coverage
  More fragments sequenced = more independent sources of evidence to call structural
variants

Experiment 5: “I have 3 patients with a rare condition and want to find the causitive
variant”
  High depth sequencing (20x?) per patient. Illumina or SOLiD or complete genomics
  Exome sequencing – 1 lane per patient
  SNPs + short indels
  Exclude all common variation (dbSNP + 1000Genomes)
  Is there a shared truncating variant? If not – is there a shared truncating structural
variant?


Q&A

Questions from you?


ECCB 2010 Next-gen sequencing Tutorial

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to ECCB 2010 Next-gen sequencing Tutorial (20)

More from Thomas Keane (14)

Recently uploaded (20)

ECCB 2010 Next-gen sequencing Tutorial