Haploid Assembly of Diploid Genomes

Challenges, Trials, Tribulations

İnanç Birol

13 October 2011

IEEE InfoVis 2009

Assembly By Short Sequencing

2

3

in Literature

• ~40 citations on tool comparisons

• ~20 citations on using ABySS for a biology study

• Crowded field – 17 teams in Assemblathon 1

4

Overlap-Overlay-Consensus

ARACHNE

CAP3

Celera assembler

MIRA

Newbler

Phred/Phrap

SGA

De Bruijn Graph

Euler

Velvet

ABySS

SOAPdenovo

ALLPATHS

Assembly Problem

A partial and unambiguous read-to-read alignment

extends the length of sequence information

• First stage of an assembly algorithm is to find such alignments

• Assembly algorithms differ in the way they find and use these alignments

5

TCGATCGATTTTCGGCCTAA read1 ATTTTCGGCCTAATATTAGG read2

…GCATCGATCGATTTTCGGCCTAATATTAGGCCGATAATCGACGATC…

Algorithm

• SE Assembly:

• PE Assembly:

• Scaffolding:

k-mer extension on a de Bruijn graph

search for unambiguous contig merging along paths

search for unambiguous linkage across distant contigs

6

d=6±5

d=5±4

d=26±9

d=12±5

Software

7

De Bruijn Graph

• Description of read-to-read overlaps

– 2x4 possible extension of every k-mer

• Provides and O(n) algorithm for SE assembly

8

…GACATTGC… seq1 …GACATTAT… seq2

GACAT ACATT

ATTAT CATTA

CATTG ATTGC

…

…

…

k = 5

Adjacency Graph

• Description of contig overlaps

– Built during SE assembly

• Overlap = k-1 bp

– Generalized during PE assembly

• Arbitrary overlap

9

Linkage Graph

• Built through read pairs aligned to different contigs

– PE reads from a tight fragment length distribution

• Reliable distance estimates

– MP reads from broader insert length distribution

• Noisy data

• Used in PE assembly (PE) and scaffolding (PE and MP) stages

10

Anchor

• Scrubbing “homozygous” variations

Indel SNPs

11

Anchor

• Local directional assembly

– scaffold gap filling (bridging)

– extension (planking)

12

Case Study

Mountain Pine Beetle Genome Assembly

13

Mountain Pine Beetle Genome

Assembly statistics

contigs scaffolds

n 1,128,463 1,103,221

n:500bp 33,591 11,657

n:N50 4,324 82

N50 (bp) 11,220 541,443

Max (bp) 276,135 3,583,207

Reconstruction (Mb) 201.9 200.4

14

Assembly As a Hairball

• ABySS v1.2.7

– PE/MP information disambiguates short contig extensions

1 2 3 4 5 6+ 1 15822 7354 1882 530 109 1

2 7354 9814 1817 456 72 3

3 1882 1817 1074 238 31 1

4 530 456 238 126 13 1 5 109 72 31 13 10 0 6+ 1 3 1 1 0 0

Node connectivity*

out in

* For contigs 2 kb

15

Scaffolding

16

Quality Assessment

Alignment of 81,047,980 reads

Gene alignments

17

Before Anchor After Anchor Change

Mapped 65,624,456 (80.97%)

66,949,341 (82.60%)

+ 1,324,885

Paired 43,207,118 (53.31%)

44,732,320 (55.19%)

+ 1,525,202

Single-end 9,536,178 (11.77%)

8,846,977 (10.92%)

-689,201

2,180 ESTs 248 Conserved Genes

Complete Partial Complete Partial

Contigs 968 1169 212 18

Scaffolds 1,481 619 228 5

Date ABySS Version

Data n:500 N50 Max Sum

August 2009 1.0.11 3x GAiix 81,431 1,526 20,755 107.3e6

November 2009 1.0.15 +2x GAiix 104,958 2,333 55,845 195.8e6

February 2010 1.1.1 +4x GAiix 157,081 2,790 136,637 346.3e6

July 2010 1.2.0 +2x GAiix 146,313 3,354 129,008 376.2e6

November 2010 1.2.4 +1x GAiix +1x GAiix

(MP)

100,690 4,474 294,323 268.8e6

May 2011 1.2.7 -- 18,660 108,158 1,908,773 201.4e6

July 2011 1.2.7 + 1x HiSeq +1x HiSeq

(MP)

11,657 541,443 3,583,207 200.4e6

August 2011 1.2.7 -- 11,523 561,847 3,746,698 206.5e6

18

Transcriptome Assembly

19

Transcriptome Sequencing

• RNA-seq protocol

• Brings information on how a genome “acts”

– Expression levels

• Allelic expression

– Present isoforms

– Gene fusions

– Other transcriptional events

– Post-transcriptional RNA editing Rodrigo Goya

20

Transcript models

Transcriptome Assembly

Transcriptome assembly is different from genome assembly

– varying coverage levels ⇒ varying expression levels

– split assembly paths ⇒ isoforms/splice variants

– small contig sizes ⇒ small product sizes

21

What Overlap to Choose?

22

Selection of k

23

What Overlap to Choose?

• Selection of parameter k depends on read coverage depth

• Expression levels vary over 5 orders of magnitude

24

Assembly Merging

25

buried parent untouched

Multi-k Assembly

We capture a wide range of expression levels

• Gray: all transcripts with a read alignment

• Blue: at least 80% of a transcript in a single contig

• Red: at least 80% of a transcript is reconstructed

26

Trans-ABySS

A versatile tool for

• Transcript reconstruction

• Gene identification

• InDel and SNV discovery

• Chimeric transcript discovery

– Gene fusions

– Trans-splicing

• Expression analysis

27

Trans-ABySS

Cufflinks 0.8.3

Scripture

28

Transcriptome Assembly

De novo assembly based on ABySS

Reference-based assembly based on TopHat alignments [Trapnell et al., 2010; Guttman et al., 2010; Trapnell et al., 2009]

Events

29 + chimeric transcripts

Performance • Compared to mapping-based analysis tools

Trans-ABySS constructs – as many transcripts

– with better sensitivity and specificity

30 [Trapnell et al., 2010; Guttman et al., 2010; Trapnell et al., 2009]

Case Study

Acute Myeloid Leukemia Transcriptome Assembly

31

Fusions • Assembled transcriptome

contigs span multiple genes

• Break point (usually) corresponds to exon boundaries

• Break point is supported by – Spanning reads – Read pairs linking regions

• Gene fusions are often drivers in AML and define subtypes (e.g. PML/RARα and M3 subtype)

1 2

4 5 6

Lucas Swanson, Readman Chiu and Gordon Robertson

32

AML Gene Fusions

0

2

4

6

8

10

12

14

16

Nu

mb

er

of

pat

ien

ts

Candidate fusion events

9%

5%

4% MLL fusions

Known AML fusion events (12) Known polymorphism (1) Novel fusion event (17)

Low frequency (<1%)

71 events in 65/173 (38%) patients 30 different gene fusions identified ≥94% validation by RT-PCR sequencing

Karen Mungall 33

Validation of a Novel Fusion

M: 1kb plus DNA ladder 1: A00160 (2938) POLR2A-FBN3

505bp

Chr 17p13.1

DNA directed RNA polymerase II polypeptide A (POLR2A)

Exon 1 2

5’UTR

Fibrillin 3 (FBN3)

Chr 19p13.2

Exon 47 48

Exon 1 5’UTR

Exon 48 Exon 63

EGF-like, calcium binding domains 1 M

Andy Mungall 34

Internal Tandem Duplications • Contig alignments result in

– Query gaps – Contiguous target blocks

• Read support on break point(s) • Aberrant read pair distances • Known AML ITDs:

– 29/173 (17%) harbour partial FLT3 exon 14 duplication

– 6/173 (3.5%) harbour partial WT1 exon 7 duplication

– Nakao et al., Leukemia 1996; Christiansen et al., Leukemia 2001 2 2’

2’

2

35

Known ITD in FLT3

• A 33 bp duplication in exon 14 CTCCCATttgagatcatattcatattctctgaaatcaacgTTGAGATCATATTCATATTCTCTGAAATCAACGTAGAA

Karen Mungall 36

Partial Tandem Duplications • Usually coexist with the wild-type • PTD event manifested in a

particular contig type – A short contig with 50/50 split

alignment

• Break point is supported by – Spanning reads – Read pairs in opposite orientation

• Known AML PTD: – 10/173 (5.8%) harbour duplication

of MLL exons 2-10 – Dorrance et al., Blood 2008

• Identified 88 genes with PTDs 2 3

37

Novel PTD in Arid1a

• Exons 2-4 tandemly repeated in 5 AML libraries

• Recurrent across tissues and species

WT CT

Source Observations

AML 5/173 Libraries

LBC 5/54 Libraries

Normal mouse 3/7 Libraries

NCBI EST colon_ins , placenta_normal

38

Summary

39

ABySS Team: Shaun Jackman Tony Raymond Rod Docking Beetle Project: Joerg Bohlmann Chris Keeling Nancy Liao Greg Taylor Simon Chan Diana Palmquist

Trans-ABySS Team: Readman Chiu Karen Mungall Gordon Robertson Ka Ming Nip Jenny Qian Rong She Lucas Swanson AML Project: Richard Moore Yongjun Zhao Andy Mungall Aly Karsan

GSC: Sequencing Team Library Core Systems Team Steven Jones Marco Marra

Final Hairball

• ABySS v1.2.7

– Read pairs and inferred distances allow for scaffolding

41

contigs scaffolds

n 1,128,463 1,103,221

n:500bp 33,591 11,657

n:N50 4,324 82

N50 (bp) 11,220 541,443

Max (bp) 276,135 3,583,207

Reconstruction (Gb) 201.9 200.4

Biotin Read-Through

circularized insert

42

43

Triage of MP Reads

Challenge: A B

A B

Which

one?

Information:

• Distances from contig ends

• Base mismatches on read ends

• Inferred contig orientations 44

Triage of MP Reads Read 1 Read 2

MP-like

PE-like

MP-like PE-like

MP-like PE-like

|x xx

|x xxx

x x|

x xxx|

|

|

45