Next-generation sequencing: informatics & software aspects

Next-generation sequencing:informatics & software aspects

Gabor T. MarthBoston College Biology Department

Harvard NanocourseOctober 7, 2009

New sequencing technologies…

… offer vast throughput … & many applications

read length

base

s per

mac

hine

run

10 bp 1,000 bp100 bp

1 Gb

100 Mb

10 Mb

10 Gb

Illumina/Solexa, AB/SOLiD sequencers

ABI capillary sequencer

Roche/454 pyrosequencer(100Mb-1Gb in 200-450bp reads)

(10-50Gb in 25-100 bp reads)

1 Mb

100 Gb

Genome resequencing for variation discovery

SNPs

short INDELs

structural variations

• the most immediate application area

Genome resequencing for mutational profiling

Organismal reference sequence

• likely to change “classical genetics” and mutational analysis

De novo genome sequencing

Lander et al. Nature 2001

• difficult problem with short reads

• promising, especially as reads get longer

Identification of protein-bound DNA

Chromatin structure (CHIP-SEQ)(Mikkelsen et al. Nature 2007)

Transcription binding sites. (Robertson et al. Nature Methods, 2007)

DNA methylation. (Meissner et al. Nature 2008)

• natural applications for next-gen. sequencers

Transcriptome sequencing: transcript discovery

Mortazavi et al. Nature Methods 2008

Ruby et al. Cell, 2006

• high-throughput, but short reads pose challenges

Transcriptome sequencing: expression profiling

Jones-Rhoads et al. PLoS Genetics, 2007

Cloonan et al. Nature Methods, 2008

• high-throughput, short-read sequencing should make a major impact, and potentially replace expression microarrays

… & enable personal genome sequencing

The re-sequencing informatics pipelineREF

(ii) read mappingIND

(i) base calling

IND(iii) SNP and short INDEL calling

(v) data viewing, hypothesis generation

(iv) SV calling GigaBayesGigaBayes

The variation discovery “toolbox”

• base callers• read mappers

• SNP callers

• SV callers

• assembly viewers

GigaBayesGigaBayes

Base error characteristics vary

Inser-tions

1.43%

Dele-tions

3.23%

Substitutions95.34%

Illumina

454

Read lengths vary

read length [bp]0 100 200 300

~200-450 (variable)

25-100(fixed)

25-50 (fixed)

25-60 (variable)

400

Sequence traces are machine-specific

Base calling is increasingly left to machine manufacturers

Representational biases

Fragment duplication

Read mapping is like a jigsaw

… and they give you the picture on the box

2. Read mapping…you get the pieces…

Unique pieces are easier to place than others…

Multiply-mapping reads

• Reads from repeats cannot be uniquely mapped back to their true region of origin

• “Traditional” repeat masking does not capture repeats at the scale of the read length

Dealing with multiple mapping

Paired-end (PE) reads

fragment length: 100 – 600bp

Korbel et al. Science 2007

fragment length: 1 – 10kb

PE reads are now the standard for whole-genome short-read sequencing

Gapped alignments (for INDELs)

The MOSAIK read mapper

• gapped mapper• option to report multiple map locations• aligns 454, Illumina, SOLiD, Helicos reads• works with standard file formats (SRF, FASTQ, SAM/BAM)

Michael Strömberg

Alignment post-processing

0 5 10 15 20 25 30 35 40 45 500

5

10

15

20

25

30

35

40

45

50

55

60

Act

ual b

ase

qual

ity

Called base quality

• quality value re-calibration

• duplicate fragment removal

Data storage requirements

Alignment visualization

• too much data – indexed browsing• too much detail – color coding, show/hide

SNP calling: old problem, new data

sequencing error polymorphism

1 2

1 21

1

1 2

Pr | Pr | Pr , , ,

Pr | Pr | Pr , , ,

Pr , , , |i

kT

ii n

l kT

nk ki i i n

i

nk k l l l li i

iG

n

B T T G G G G

B T T G G G G

G G G B

Allele calling in next-gen data

SNP

INS

New technologies are perfectly suitable for accurate SNP calling, and some also for short-INDEL detection

SNP calling in multi-sample read sets

P(G1=aa|B1=aacc; Bi=aaaac; Bn= cccc)P(G1=cc|B1=aacc; Bi=aaaac; Bn= cccc)P(G1=ac|B1=aacc; Bi=aaaac; Bn= cccc)

P(Gi=aa|B1=aacc; Bi=aaaac; Bn= cccc)P(Gi=cc|B1=aacc; Bi=aaaac; Bn= cccc)P(Gi=ac|B1=aacc; Bi=aaaac; Bn= cccc)

P(Gn=aa|B1=aacc; Bi=aaaac; Bn= cccc)P(Gn=cc|B1=aacc; Bi=aaaac; Bn= cccc)P(Gn=ac|B1=aacc; Bi=aaaac; Bn= cccc)

P(SNP)

“genotype probabilities”

P(B1=aacc|G1=aa)P(B1=aacc|G1=cc)P(B1=aacc|G1=ac)

P(Bi=aaaac|Gi=aa)P(Bi=aaaac|Gi=cc)P(Bi=aaaac|Gi=ac)

P(Bn=cccc|Gn=aa)P(Bn=cccc|Gn=cc)P(Bn=cccc|Gn=ac)

“genotype likelihoods”

Prio

r(G1,.

.,Gi,..

, Gn)

-----a----------a----------c----------c-----

-----a----------a----------a----------a----------c-----

-----c----------c----------c----------c-----

Trio sequencing

2

2

2 22 2

2

2

2

2 2

2

11 12 22

1 111: 1 12 2 11: 111: 1

1 111 12 : 2 1 12 : 2 1 1 12 : 12 2

22 : 22 : 11 122 : 12 2

1 1 111: 1 1 11:2 2 4

Pr | , 1 112 12 : 2 1 12 2

1 122 : 12 2

M M M

F

C M FF

G G G

G

G G GG

2 2 2

2 22 2

2 22

2

2 22 2

1 1 1 11 1 11: 12 4 2 2

1 1 1 1 112 : 2 1 1 2 1 12 : 1 2 14 2 4 2 2

1 1 1 1 122 : 1 1 22 : 1 14 2 4 2 2

1 111: 12 211: 1

1 122 12 : 1 12 : 12

22 : 1FG

2

22

11:2 1 12 : 2 1

222 : 11 122 : 1 1

2 2

• the child inherits one chromosome from each parent• there is a small probability for a de novo (germ-line or somatic) mutation in the child

Alignment visualization

• too much data – indexed browsing• too much detail – color coding, show/hide

Standard data formats

SRF/FASTQ

SAM/BAM

GLF/VCF

Human genome polymorphism projects

common SNPs

Human genome polymorphism discovery

The 1000 Genomes Project

Pilot 1

Pilot 2

Pilot 3

1000G Pilot 3 – exon sequencing

• Targets:1K genes / 10K targets

• Capture: Solid / liquid phase

• Sequencing:454 / IlluminaSE / PE

• Data producers:BaylorBroadSangerWash. U.

• Informatics methods:Multiple read mapping &SNP calling programs

Coverage varies

On/off target capture

ref allele*:45%

non-ref allele*: 54%Target region

SNP(outside target region)

Fragment duplication – revisited

Reference allele bias

(*) measured at 450 het HapMap 3 sites overlapping capture target regions in sample NA07346

ref allele*:54%

non-ref allele*: 45%

SNP calling findings

BCM/454 BI/SLX WUGSC/SLX SC/SLX# Samples 32 23 16 11

<read depth> per sample 35 X 62 X 117 X 51 X

# SNPs called 7,200 – 8,400 4,500 – 4,700 3,700 3,500 – 3,700

% dbSNPs 39 - 55 65 - 72 68 75 - 85

Ts/Tv(#SNP) 1.7 – 2.6 1.9 – 2.3 2.3 2.5 – 2.6

# Novel SNPs 3,998 1,550 1,947 892

• based on a method comparison / testing exercise

• 80 samples drawn from the 4 Centers

• read mapping / SNP calling by the Baylor pipeline (BCM/454 data); the Broad and the BC pipelines (all 80 samples)

Overlap between call sets

45217238.05%1.21

2,2961,86281.10%3.40

413245.81%0.23

Broad callsBC calls

# SNP calls:# dbSNPs:% dbSNPs:Ts/Tv ratio:

The 1000G Structural Variation Discovery Effort

Structural variation detection

Feuk et al. Nature Reviews Genetics, 2006

SV detection – resolution

Expected CNVsKaryotype

Micro-arraySequencing

Rela

tive

num

bers

of e

vent

s

CNV event length [bp]

Detection Approaches

• Read Depth: good for big CNVs

Sample Reference

Lmap

read

contig

• Paired-end: all types of SV

• Split-Readsgood break-point resolution

• deNovo Assembly~ the future

SV slides courtesy of Chip Stewart, Boston College

47

Read depth (RD)

Statistical & systematic biases

Single molecule sequencing?

GC Bias

Coverage bias

RD resolution

dens

ity (l

og10

)

Illum

ina

obse

rved

read

cou

nts (

per k

b)

expected read counts ( per kb)

CN = 2

CN = 1

CN = 3

CNV events detected with RD

Read

s/2k

b lo

g 2(o

/e)

Chromosome 2 Position [Mb]

Read

s/5k

b lo

g 2(o

/e)

Read

s/kb

log 2

(o/e

)

Helicos

40 kb deletion 454

Illumina

individual “NA12878”

SV detection with PE read map positions

Deletion

Individual Reference

LMLdel

Tandemduplication Ldup

InversionLinv

Fragment length distributions

• long fragments ~ better fragment coverage and sensitivity to large events (454)• tighter distributions ~ better breakpoint resolution and sensitivity for shorter events (Illumina)

The SV/CNV event display

chromosomeoverview

fragment lengths

read depth

eventtrack

300 bp deletion in chromosome of NA12878 by Illumina paired-end data from the 1000 Genomes project

Chip Stewart

Deletion event lengths

ALUY

L1

Mobile element insertions: PE reads

• Used with short-read data (Illumina, in our case)• Detect clusters of 5’ & 3’ read pairs with one end mapping

to a mobile element• Clusters far from annotated elements are candidate

insertion events

ALU / L1 / SVA5’ end 3’ end

Mobile element insertions: Split reads

• Requires longer reads (454)• Reads “mapping into” mobile element not present in the

reference genome sequence are candidate insertion events

Mobile element insertions: trio members

356185 182

NA12891684

NA12892664

ALU insertions

NA12878733

47

7996

10

Detection in 1000 G Pilot 2 CEU trio PE Illumina data

Mobile element insertions: PE vs. Split-reads

569247 163454 Split-Read817

SLX PE733

ALU insertions in NA12878

BC event lists in 1000 Genomes data

SV typePilot 1140

samples low coverage

Pilot 26 samples

high coverage

deletions 5,555 4,718tandem

duplications 540 406mobile

element insertions

3,276 2,013

SV calls / validation in 1000G datasets

Software access

http://bioinformatics.bc.edu/marthlab/Software_Release

Credits

Elaine Mardis

Andy Clark

Aravinda Chakravarti

Michael Egholm

Scott Kahn

Francisco de la Vega

Patrice MilosJohn Thompson

R01 HG004719R01 HG003698R21 AI081220RC2 HG5552

Lab

Several positions are available:grad students / postodocs /

programmers

Can we find exon CNVs in Pilot 3 data?

SVs in exon sequencing data