9/7/2013

1

Análise

Genômica

Guilherme T Valente (valentegt@gmail.com)Centro de Biologia e Engenharia GenéticaUniversidade Estadual de Campinas - UNICAMPCampinas, SP

Uma overview sobre processamento de high throughput sequencing

9/7/2013

2

Sequenciamento de ácidos nucleicos Processo que

determina a sequência de nucleotídeos de uma porção do

genoma ou de um genoma completo

Método de Maxam-Gilbert (1977) Método de Sanger (1977)

Primeiros Métodos First generation sequence

Géis e capilar

9/7/2013

3

Método de Sanger é o mais difundido até os dias atuais

1989 Primeiro gene humano (proteína mutada gera a fibrose cística);

1995 O genoma de Haemophilus influenzae é o primeiro genoma a ser sequenciado;

1996 O genoma de Saccharomyces cerevisiae é o primeiro genoma eucarionte a ser

sequenciado;

1998 O genoma de C. elegans é o primeiro genoma de um pluricelular a ser

sequenciado;

2001 - Primeiro rascunho da sequência do genoma humano é publicado;

2003 - 99% do genoma humano foi sequenciado pelo Projeto do Genoma Humano.

Métodos Atuais The next generation sequence

1 - melhor custo-benefício para projetos high throughput sequencing;

2 custo por bp muito menor;

3 sequenciamento muito mais rápido e eficiente (>40 Gbase/corrida).

4 versatilidade Genomas, expressão gênica, diagnóstico, gen. população,

epigenética, metagenômica, entre outros

9/7/2013

4

Métodos Atuais The next generation sequence

Plataf. Comp

(bp)

Tempo de

corrida

(dias)

Gb por corrida Prós Contras

FLX Titanium

330 0,35 0,45 Reads longos;

agilidade

Alto custo dos

reagentes; alto erro

Illumina/Solexa 75-

~100

9 35 Plataforma mais

utilizada

Baixa capacidade

multiplexar

amostras

SOLiD 3 50 14 50 Sistema de correção

de erros

Demora na corrida

Helicos BioSciences

HeliScope

32 8 37 Sem viés por

representação do

template pós amplific.

Alto erro

Polonator G.007 26 5 12 Menor custo; química

aberta para adapt.

Menores reads

Metzker 2010

* HiSeq2000 gera um output com 600Gbp por corrida. Maior output atual.

5.292,39

0,09

-500,00

500,00

1.500,00

2.500,00

3.500,00

4.500,00

5.500,00

Sep

tem

ber

-20

01

Mar

ch-2

00

2

Sep

tem

ber

-20

02

Mar

ch-2

00

3

Oct

ob

er-2

00

3

Jan

uar

y-2

00

4

Ap

ril-

20

04

July

-20

04

Oct

ob

er-2

00

4

Jan

uar

y-2

00

5

Ap

ril-

20

05

July

-20

05

Oct

ob

er-2

00

5

Jan

uar

y-2

00

6

Ap

ril-

20

06

July

-20

06

Oct

ob

er-2

00

6

Jan

uar

y-2

00

7

Ap

ril-

20

07

July

-20

07

Oct

ob

er-2

00

7

Jan

uar

y-2

00

8

Ap

ril-

20

08

July

-20

08

Oct

ob

er-2

00

8

Jan

uar

y-2

00

9

Ap

ril-

20

09

July

-20

09

Oct

ob

er-2

00

9

Jan

uar

y-2

01

0

Ap

ril-

20

10

July

-20

10

Oct

ob

er-2

01

0

Jan

uar

y-2

01

1

Ap

ril-

20

11

July

-20

11

Jan

-20

12

(EST

)

http://genome.gov/splash.htm

CUSTO DO SEQUENCIAMENTO POR BASE

Roche 454 Illumina

Solid e Helicos

9/7/2013

5

95.263.072

7.950

0

10.000.000

20.000.000

30.000.000

40.000.000

50.000.000

60.000.000

70.000.000

80.000.000

90.000.000

100.000.000

Sep

tem

ber

-20

01

Mar

ch-2

00

2

Sep

tem

ber

-20

02

Mar

ch-2

00

3

Oct

ob

er-2

00

3

Jan

uar

y-2

00

4

Ap

ril-

20

04

July

-20

04

Oct

ob

er-2

00

4

Jan

uar

y-2

00

5

Ap

ril-

20

05

July

-20

05

Oct

ob

er-2

00

5

Jan

uar

y-2

00

6

Ap

ril-

20

06

July

-20

06

Oct

ob

er-2

00

6

Jan

uar

y-2

00

7

Ap

ril-

20

07

July

-20

07

Oct

ob

er-2

00

7

Jan

uar

y-2

00

8

Ap

ril-

20

08

July

-20

08

Oct

ob

er-2

00

8

Jan

uar

y-2

00

9

Ap

ril-

20

09

July

-20

09

Oct

ob

er-2

00

9

Jan

uar

y-2

01

0

Ap

ril-

20

10

July

-20

10

Oct

ob

er-2

01

0

Jan

uar

y-2

01

1

Ap

ril-

20

11

July

-20

11

Jan

-20

12

(EST

)

Genoma humano U$ 3 bi; ~3 bilhões bp; 11 anos; ~12X

Genoma U$ 2.000,00; ~1 bilhão bp; alguns dias; ~30X

CUSTO DO SEQUENCIAMENTO POR GENOMA

http://genome.gov/splash.htm

Roche 454 Illumina

Solid e Helicos

Illumina HiSeq: 600 Gb

9/7/2013

6

Genoma humano - ~3 bilhões = ~800 Alberts, 10

anos

Genoma de um ciclídeo - ~1 bilhão x 30 = ~30

bilhões = ~8.000 Alberts, alguns dias

Antes e hoje

Objetivo rastrear as variações genéticas do genoma humano.

Objetivo gerar conhecimento sobre o genoma de câncer com o intuito de gerar

tratamentos e diagnósticos

Hoje

9/7/2013

7

Métodos do futuro próximo The third generation sequence

* - Perspectiva para os próximos anos. Promete sequenciar um genoma humano em 15 min.

Vantagens ao NGS sem PCR; reads longos

Plataf. Comp

(bp)

Tempo de

corrida

(dias)

Gb por corrida Prós Contras

Pacific

Biosciences

(target release:

2010)

3.000

(alguns

>20.000)

? ? Reads mais longos Alto erro

NanoPore *100.000 *15 min ? Reads mais longos Alto erro

Metzker 2010; Hayden 2012

-time sequencing technology (SMRT)

9/7/2013

8

History of Genome Sequencing The Sanger Method

The sequencing of DNA molecules began in the 1970s with development of the Maxam-Gilbert

ethod, and later the Sanger method. Originally developed by Frederick Sanger in 1975, most

DNA sequencing that occurs in medical and research laboratories today is performed using

sequencers employing variations of the Sanger method. Termed the chain-termination method, it

involves a reaction where chain-terminator nucleotides are labeled with fluorescent dyes,

combined with fragmented DNA, DNA sequencing primers and DNA polymerase. Each

nucleotide in the DNA sequence is labeled with a different dye color and a chromatogram is

produced, with each color representing a different letter in the DNA code A, T, C, or G.

Advances in sequencing technology and computer programming enabled relatively fast and cost

efficient DNA sequencing. However, sequencing of entire genomes of organisms was difficult

and time consuming. At the time the Human Genome Project was officially started in 1990, it was

thought that sequencing the human genome would take fifteen years. The sequence was

released in 2003, although some gaps still exist. The estimated project cost for the entire Human

Genome Project is around $3 billion, although this figure represents a wide range of scientific

activities that went into the project, not exclusively genome sequencing. Optimally, current

sequencers are able to sequence approximately 2.8 million base pairs per 24 hour period.

However, even the smallest organisms such as bacteria are hundreds of millions of base-pairs

long and the human genome is about 3 billion (3,000,000,000) base pairs. At this rate, using the

most modern Sanger sequencers, it takes almost three years to sequence the human genome.

History of Genome Sequencing

Discovery of DNA structure by Watson and Crick (1953)

1950 1960 1970 1980 1990 2000

Development of Sanger Sequencing by Frederick

Sanger (1977)

Polymerase Chain Reaction (PCR) invented by Kary Mullis

(1983) Human Genome

Project Initiated

454 Life Sciences Corp. is

founded

Sequence of Human Genome Completed

(2003)

Development of Pyrosequencing by

454 Sequencing was used to sequence the following genomes (along with many others): - Barley (plant) - Neanderthal - Helicobacter pylori (H.pylori - bacteria that cause gastric ulcers) - HIV mutant strains

Source: U.S. Department of Energy Office of Science, Systems Biology for Energy and the Environment, Human Genome Project Information Base URL: http://genomics.energy.gov/

2012-2013

Reads >10 kbp

Kahn SD. Science 2011

9/7/2013

9

Genoma

referência

NGS

Reads

Análise dos reads

De novo

DNA/RNA

Bibliotecas

9/7/2013

10

Maiores

problemas dos

sequenciamentos

em larga escala!!!Estocagem

Montagem

Análise

Estocagem

9/7/2013

11

Estocagem

Genoma de referência De novo assemblyX

1 - Perda de cópias de genes surgidos recentementes.

2 Algumas reg. podem não existir no gen. ref.

3 - Problemas na montagem dos reads referentes a elementos

repetitivos (problemas principal nos eucariontes).

Montagem

9/7/2013

12

Análise X- Não existe um programa aplicável a responder todas as perguntas biológicas;

- Programas novos emergem diariamente.

Genoma

referência

NGS

Reads

Análise dos reads

De novo

DNA/RNA

Bibliotecas

9/7/2013

13

Utilizando um genoma de referência

Short-read mappers

- Muito utilizado atualmente nos projetos genomas e transcriptomas;

- Recomendado quando há uma boa referência (genoma ou transcriptoma).

NATURE BIOTECHNO LO GY VOLUME 27 NUMBER 5 MAY 2009 455

to understand why the mapping problems are

computationally difficult, which difficulties

have been overcome and what challenges and

opportunities remain.

Challenges of mapping short reads

The first challenge is a practical one: if the

reference genome is very large, and if we have

billions of reads, how quickly can we align the

reads to the genome? Sequence alignment is a

classic problem in bioinformatics, supported

by a large body of literature describing different

variants for both exact and inexact alignment.

As a practical matter, the task of mapping

billions of sequences to a mammalian-sized

genome calls for extraordinarily efficient algo-

rithms, in which every bit of memory is used

optimally or near optimally.

The second challenge is strategic: if a read

comes from a repetitive element in the refer-

ence, a program must pick which copy of the

repeat the read belongs to. Because this may

be impossible to decide with confidence, the

program may choose to report multiple pos-

sible locations or to pick a location heuristi-

cally. Sequencing errors or variations between

the sequenced chromosomes and the reference

genome exacerbate this problem, because the

In this case, to make sense of the reads, their

positions within the reference sequence must

be determined. This process is known as align-

ing or mapping the read to the reference. In

one version of the mapping problem, reads

must be aligned without allowing large gaps in

the alignment (we describe this in more detail

in the Short-read mappers section below).

A more difficult version of the problem arises

primarily in RNA-Seq, in which alignments are

allowed to have large gaps corresponding to

introns (discussed below in the Spliced-read

mappers section).

These read mapping problems are certainly

not new, and there are many programs that

perform both spliced and unspliced alignment

for the older Sanger-style capillary reads. Even

so, these programs neither scale up to the much

greater volumes of data produced by short-

read sequencers nor scale down to the short

read lengths. Aligning the reads from ChIP-Seq

or RNA-Seq experiments can take hundreds or

thousands of central processing unit (CPU)

hours using conventional software tools such

as BLAST or BLAT. Fortunately, new software

packages designed to meet the computational

challenges of short-read sequencing are quickly

appearing. Before choosing one, it is essential

A new generation of DNA sequencers that can

rapidly and inexpensively sequence billions

of bases is transforming genomic science. These

new machines are quickly becoming the tech-

nology of choice for whole-genome sequencing

and for a variety of sequencing-based assays,

including gene expression, DNA-protein inter-

action, human resequencing and RNA splicing

studies . For example, the RNA-Seq proto-

col, in which processed mRNA is converted to

cDNA and sequenced, is enabling the identifi-

cation of previously unknown genes and alter-

native splice variants; the ChIP-Seq approach,

which sequences immunoprecipitated DNA

fragments bound to proteins, is revealing net-

works of interactions between transcription

factors and DNA regulatory elements4; and

the whole-genome sequencing of tumor cells

is uncovering previously unidentified cancer-

initiating mutations5.

One of the challenges presented by the new

sequencing technology is the so-called read

mapping problem. Sequencing machines made

by Illumina of San Diego, Applied Biosystems

(ABI) of Carlsbad, California, and Helicos

of Cambridge, Massachusetts, produce short

sequences of 25 100 base pairs (bp), called

reads, which are sequence fragments read

from a longer DNA molecule present in the

sample that is fed into the machine. In con-

trast to whole-genome assembly, in which these

reads are assembled together to reconstruct a

previously unknown genome, many of the

next-generation sequencing projects begin

with a known, or so-called reference, genome.

How to map billions of short reads onto genomesCole Trapnell & Steven L Salzberg

Mapping the vast quantities of short sequence fragments produced by next-generation sequencing platforms is a

challenge. What programs are available and how do they work?

Cole Trapnell and Steven L. Salzberg are at the

Center for Bioinformatics and Computational

Biology, University of Maryland, College Park,

Maryland, USA.

e-mail: cole@cs.umd.edu or salzberg@umd.edu

Table 1 A selection of short-read analysis software

Program Website

Open

source?

Handles ABI color

space?

Maximum read

length

Bowtie http://bowtie.cbcb.umd.edu Yes No None

BWA http://maq.sourceforge.net/bwa-man.shtm l Yes Yes None

Maq http://maq.sourceforge.net Yes Yes 127

Mosaik http://bioinformatics.bc.edu/marthlab/Mosai k No Yes None

Novoalign http://www.novocraft.com No No None

SOAP2 http://soap.genomics.org.c n No No 60

ZOOM http://www.bioinfor.com No Yes 240

PRI M ER

NATURE BIOTECHNO LO GY VOLUME 27 NUMBER 5 MAY 2009 455

to understand why the mapping problems are

computationally difficult, which difficulties

have been overcome and what challenges and

opportunities remain.

Challenges of mapping short reads

The first challenge is a practical one: if the

reference genome is very large, and if we have

billions of reads, how quickly can we align the

reads to the genome? Sequence alignment is a

classic problem in bioinformatics, supported

by a large body of literature describing different

variants for both exact and inexact alignment.

As a practical matter, the task of mapping

billions of sequences to a mammalian-sized

genome calls for extraordinarily efficient algo-

rithms, in which every bit of memory is used

optimally or near optimally.

The second challenge is strategic: if a read

comes from a repetitive element in the refer-

ence, a program must pick which copy of the

repeat the read belongs to. Because this may

be impossible to decide with confidence, the

program may choose to report multiple pos-

sible locations or to pick a location heuristi-

cally. Sequencing errors or variations between

the sequenced chromosomes and the reference

genome exacerbate this problem, because the

In this case, to make sense of the reads, their

positions within the reference sequence must

be determined. This process is known as align-

ing or mapping the read to the reference. In

one version of the mapping problem, reads

must be aligned without allowing large gaps in

the alignment (we describe this in more detail

in the Short-read mappers section below).

A more difficult version of the problem arises

primarily in RNA-Seq, in which alignments are

allowed to have large gaps corresponding to

introns (discussed below in the Spliced-read

mappers section).

These read mapping problems are certainly

not new, and there are many programs that

perform both spliced and unspliced alignment

for the older Sanger-style capillary reads. Even

so, these programs neither scale up to the much

greater volumes of data produced by short-

read sequencers nor scale down to the short

read lengths. Aligning the reads from ChIP-Seq

or RNA-Seq experiments can take hundreds or

thousands of central processing unit (CPU)

hours using conventional software tools such

as BLAST or BLAT. Fortunately, new software

packages designed to meet the computational

challenges of short-read sequencing are quickly

appearing. Before choosing one, it is essential

A new generation of DNA sequencers that can

rapidly and inexpensively sequence billions

of bases is transforming genomic science. These

new machines are quickly becoming the tech-

nology of choice for whole-genome sequencing

and for a variety of sequencing-based assays,

including gene expression, DNA-protein inter-

action, human resequencing and RNA splicing

studies . For example, the RNA-Seq proto-

col, in which processed mRNA is converted to

cDNA and sequenced, is enabling the identifi-

cation of previously unknown genes and alter-

native splice variants; the ChIP-Seq approach,

which sequences immunoprecipitated DNA

fragments bound to proteins, is revealing net-

works of interactions between transcription

factors and DNA regulatory elements4; and

the whole-genome sequencing of tumor cells

is uncovering previously unidentified cancer-

initiating mutations5.

One of the challenges presented by the new

sequencing technology is the so-called read

mapping problem. Sequencing machines made

by Illumina of San Diego, Applied Biosystems

(ABI) of Carlsbad, California, and Helicos

of Cambridge, Massachusetts, produce short

sequences of 25 100 base pairs (bp), called

reads, which are sequence fragments read

from a longer DNA molecule present in the

sample that is fed into the machine. In con-

trast to whole-genome assembly, in which these

reads are assembled together to reconstruct a

previously unknown genome, many of the

next-generation sequencing projects begin

with a known, or so-called reference, genome.

How to map billions of short reads onto genomesCole Trapnell & Steven L Salzberg

Mapping the vast quantities of short sequence fragments produced by next-generation sequencing platforms is a

challenge. What programs are available and how do they work?

Cole Trapnell and Steven L. Salzberg are at the

Center for Bioinformatics and Computational

Biology, University of Maryland, College Park,

Maryland, USA.

e-mail: cole@cs.umd.edu or salzberg@umd.edu

Table 1 A selection of short-read analysis software

Program Website

Open

source?

Handles ABI color

space?

Maximum read

length

Bowtie http://bowtie.cbcb.umd.edu Yes No None

BWA http://maq.sourceforge.net/bwa-man.shtm l Yes Yes None

Maq http://maq.sourceforge.net Yes Yes 127

Mosaik http://bioinformatics.bc.edu/marthlab/Mosai k No Yes None

Novoalign http://www.novocraft.com No No None

SOAP2 http://soap.genomics.org.c n No No 60

ZOOM http://www.bioinfor.com No Yes 240

PRI M ER

Qual programa?

9/7/2013

14

Algoritmo indexing

1 Reads são fragmentados (seed);

2 Compara dois pares de seeds por vez contra

a referência.

Algoritmo indexing

1 O genoma ref. é concatenado;

2 Alinhamento de cada base por vez. A cada

ciclo, o número de possíveis regiões de

alinhamento é reduzido.

9/7/2013

15

456 VOLUME 27 NUMBER 5 MAY 2009 NATURE BIOTECHNO LO GY

Short-read mappers

Such programs as Maq and Bowtie (Table 1)

use a computational strategy known as index-

ing to speed up their mapping algorithms. Like

the index at the end of a book, an index of a

large DNA sequence allows one to rapidly find

shorter sequences embedded within it. Maq is

based on a straightforward but effective strategy

called spaced seed indexing6 (Fig. 1a). In this

strategy, a read is divided into four segments of

equal length, called the seeds. If the entire read

aligns perfectly to the reference genome, then

clearly all of the seeds will also align perfectly.

If there is one mismatch, however, perhaps due

to a single-nucleotide polymorphism (SNP),

then it must fall within one of the four seeds,

but the other three will still match perfectly.

Using similar reasoning, two mismatches will

fall in at most two seeds, leaving the other two

to match perfectly. Thus, by aligning all pos-

sible pairs of seeds (six possible pairs) against

the reference, it is possible to winnow the list of

candidate locations within the reference where

the full read may map, allowing at most two

mismatches. Maqs spaced seed index enables

it to perform this winnowing operation very

efficiently. The resulting set of candidate reads

is typically small enough that the rest of the

read that is, the other two seeds that might

contain the mismatches may be individually

checked against the reference.

Bowtie takes an entirely different approach,

borrowing a technique originally developed

for compressing large files called the Burrows-

Wheeler transform. Using this transform, the

index for the entire human genome fits into

less than two gigabytes of memory (an amount

that is commonly available on todays desktop

and even laptop computers) in contrast to a

spaced seed index, which may require over 50

gigabytes and yet reads can still be aligned

efficiently. Bowtie aligns a read one character

at a time to the Burrows-Wheeler transformed

genome (Fig. 1b). Each successively aligned

new character allows Bowtie to winnow the

list of positions to which the read might map.

If Bowtie cannot find a location where a read

aligns perfectly, the algorithm backtracks

to a previous character of the read, makes a

substitution and resumes the search. In effect,

the Burrows-Wheeler transform enables

Bowtie to conquer the mapping problem by

first solving a simple subproblem align one

character and then building on that solution

to solve a slightly harder problem align two

characters and then continuing on to three

characters, and so on, until the entire read has

been aligned. Bowties alignment algorithm is

substantially more complicated than Maqs, but

Bowties alignment speed is more than 30-fold

faster7.

using traditional alignment algorithms such as

BLAST or BLAT, but such grids are not acces-

sible to everyone. To reduce the computing cost

of analysis for sequencing-based assays and to

make them available to all investigators, we and

others have created a new generation of align-

ment programs capable of mapping hundreds

of millions of short reads on a single desktop

computer. Vendors of sequencing machines

provide specialized mapping software, such as

the ELAND program from Illumina, but in this

article we focus on third-party packages, some of

which are free and open source. These programs

are built on algorithms that exploit features of

short DNA sequencing reads to map millions of

reads per hour while minimizing both process-

ing time and memory requirements.

alignment between the read and its true source

in the genome may actually have more differ-

ences than the alignment between the read

and some other copy of the repeat. The spliced

mapping problem faces this same challenge but

is further complicated by the possible presence

of intron-sized gaps.

DNA sequencers from Illumina, ABI, Roche

(of Basel, Switzerland), Helicos and other compa-

nies produce millions of reads per run. Complete

assays may involve many runs, so an investigator

may need to map millions or billions of reads

to a genome. For example, the recent cancer

genome sequencing project by Ley et al.5 gener-

ated nearly 8 billion reads from 132 sequencing

runs. A large, expensive computer grid might

map the reads from this experiment in a few days

Burrows-Wheeler

Reference genome(> 3 gigabases)

Bowtie index(~2 gigabytes)

Concatenate intosingle str ing

Burrows-Wheelertransform and indexing

ACTCCCGTACTCTAAT

AAT

AT

T

ACTCCCGTACTCTAAT

Convert eachhit back togenome location

Position N

Position 2

Index seed pairs

Position 1

Seed index(tens of gigabytes)

Hits identify positionsin genome wherespaced seed pairis found

Look up each pairof seeds in index

Six seedpairs perread/fragment

Confirm hitsby checking

**** positions

ACTC CCGT ACTC TAAT

Report alignment to user

Chr1Chr2Chr3Chr4

Spaced seeds

Reference genome(> 3 gigabases)

Short read

Extract seeds

ACTCCCGTACTCTAAT

Short read

ACTCCCGTACTCTAAT

Look up

of read

Hits identifypositions in

genome whereread is found

1

2

3

4

5

6

CTGC CGTA AACT AATG

ACTG CCGT AAAC TAAT

ACTG **** AAAC ****

ACTG **** AAAC ****

**** CCGT **** TAAT

ACTG **** **** TAAT

**** **** AAAC TAAT

ACTG CCGT **** ****

**** CCGT AAAC ****

**** CCGT **** TAAT

ACTG **** **** TAAT

**** CCGT AAAC ****

ba

Chr1Chr2Chr3Chr4

Figure 1 Two recent algorithmic approaches for aligning short (20 200-bp) sequencing reads.

(a) Algorithms based on spaced-seed indexing, such as Maq, index the reads as follows: each position

in the reference is cut into equal-sized pieces, called seeds and these seeds are paired and stored

in a lookup table. Each read is also cut up according to this scheme, and pairs of seeds are used as

keys to look up matching positions in the reference. Because seed indices can be very large, some

algorithms (including Maq) index the reads in batches and treat substrings of the reference as queries.

(b) Algorithms based on the Burrows-Wheeler transform, such as Bowtie, store a memory-efficient

representation of the reference genome. Reads are aligned character by character from right to left

against the transformed string. With each new character, the algorithm updates an interval (indicated

by blue beams ) in the transformed string. When all characters in the read have been processed,

alignments are represented by any positions within the interval. Burrows-Wheeler based algorithms can

run substantially faster than spaced seed approaches, primarily owing to the memory efficiency of the

Burrows-Wheeler search. Chr., chromosome.

PRI M ER

Alinha 70-75% dos reads

Desafios:

1 Alinhamento de indels;

2 Ajustamento dos parâmetros;

3 - Mapeamento correto dos TEs ou outros DNAs repetitivos: a cópia

alinhada é realmente pertencente aquela posição?

4 Péssima escolha se não houver uma referência com qualidade.

Problemas?

9/7/2013

16

Montagem de genoma - de novo assembly

Algoritmos Prós Contras

Overlaps Mais sensíveis Mais complexos

Shared K-mers Menos complexos Menos sensíveis

Assemblies de NGS usando teoria de grafos

overlap-layout-consensus assemblers

De Bruijn graph assemblers

Greedy assemblers

Vértices (reads)

Arestas (overlapping)

9/7/2013

17

Leonard Euler 1736

Encontrar um caminho pela cidade que cruze cada ponte somente uma vez

Teoria de grafos(redes)

33

Problema matemático - As sete pontes de Königsberg (ex-capital da Prussia)

O algoritmo baseado em overlaps-layout-consensus (OLC) compara todos os reads

entre si

Vértices (reads)

Arestas (overlapping)

1 Todos os reads são comparados entre si;

2 Reads que se sobrepõem são selecionados;

3 A sequencia consenso é determinada.

9/7/2013

18

O algoritmo baseado em greedy assemblers alinha par-a-par todos os reads e os dois

maiores overlapping são escolhidos.

Algoritmo baseado em De Bruijn graph. Nós representam prefixos e sulfixos de

um alfabeto finito, enquanto as arestas são os overlapping entre essas letras

9/7/2013

19

De Bruijn graph é baseado no problema das sete pontes de Königsberg .

A alta exigência de RAM é devido ao tamanho do grafo.

9/7/2013

20

As sequências repetitivas introduzem um grande número de caminhos válidos

RAM, tempo e acurácia.

A Practical Compariso

n of De Novo Genome Assembly

Software Tools for Next-Generation Sequencing

Technologies

Wenyu Zhang, JiajiaChen, Yang Yang, Yifei Tang, Jin

g Shang, Bairong Shen*

Center for Systems Biology, Soochow University, Suzhou, Jiangsu, China

Abst ract

The advent of next-generation sequencing technologies is accompanied with

the development of many whole-genome

sequence assembly methods and software, especially for denovo fragment assembly. Due to the poor knowledge about the

applicability

and performanceof these softw

are tools, choosing a befitting assembler becomes a tough task. Here, we

provide the information of adaptivityfor each

program, then above all, compare the performanceof eight distin

ct tools

against eight groups of simulated datasets from

Solexa sequencing platform. Considering the computational time,

maximum random access memory (RAM) occu

pancy, assembly accuracy and integrity, our study indicate that strin

g-based

assemblers,overlap-layout-co

nsensus (OLC) assemblers are well-suited for very short reads and longer reads of small

genomes respectively. For large datasets of more than hundred millio

ns of short reads, De Bruijn graph-based assemblers

would be more appropriate. In terms of software implementation, strin

g-based assemblers are superior to graph-based

ones, of whichSOAPdenovo is complex for the creation of configuration file. Our compariso

n study will assist researchers in

selecting a well-su

ited assembler and offer essential information for the improvement of existing assemblers or the

developing of novel assemblers.

Citat ion: Zhang W, Chen J, Yang Y, Tang Y, Shang J, et al. (2011) A Practical Compariso

n of De Novo Genome Assembly Software Tools for Next-Generation

Sequencing Technologies. PLoS ONE 6(3): e17915. doi:10.1371/journal.pone.0017915

Editor: I. King Jordan, Georgia Institute of Technology, United States of America

Received November 20, 2010; Accept ed February 15, 2011; Published March14, 2011

Copyright :2011 Zhang et al. This is an open-acce

ss article distri

buted under the terms of the Creative Commons Attribution License, which

permits

unrestricted use, distri

bution, and reproduction in any medium, provided the original author and source

are credited.

Funding: This work was supported by grants from the National Nature ScienceFoundation of China (31040020) and the National 973 Programs of China

(2007CB947002, 2010CB945600). The funders had no role in study design, data collection and analysis,

decision to publish

, or preparation of the manuscript.

Compet ing Interests: The authors have declared that no competing interestsexist.

* E-mail: bairong.shen@suda.edu.cn

Introduct ion

Inrece

nt years,the next-generation sequencing (or deep

sequencing) technologies

have beenevolving rapidly, with

the

potential toaccel

eratebiological and

biomedical research

dramatically

[1]. However,the downstre

amanalysis

of short

reads datasetsafter

sequencing is a tough task; one of the biggest

challenges for the analysis of high throughput sequencing reads is

the whole genome assembly. DNA fragment assem

bly has a long

history since the emergence of the first

generation of sequencing

technologies

[2,3]. The assembly proced

ure becomes especia

lly

difficult when

tackling short and high throughput reads with

different erro

r profiles[4]. Acco

rding to the existence of refe

rence

information, theassembly proced

ure can beclassified asrefe

rence-

guidegenome assembly and denovo

genome assembly, of which

the

former is relatively

toilless with

the aid of reference genome or

proteome information while the laterin more challen

ging. Herein,

we focus on the comparison and evaluation of tools for de novo

assembly of genome sequence.

The genome assemblersgenerally take a file

of short sequence

reads and a fileof quality-value as the input. Since the quality-

value filefor the high throughput short reads is usually

highly

memory-intensive, onlya

fewassemblers

, forexample,

SHARCGS [5], and ALLPATHS-LG[6] adopt it in

the

posterior assembly

process. For the sake of computational

memory saving and convenience of data inquiry, high-through-

put short reads data is always initially formattedto specif

ic data

structure. Curren

tly, existing data stru

cture for this usage can be

predominantly

classified

into two categories: strin

g-basedmodel

and graph-basedmodel. We theref

orecall the corresponding

assemblers as strin

g-basedand graph-based. Strin

g-basedassem

-

blers, implemented with

Greedy-extension algorithm, are mainly

reported for the assem

bly of small genomes [5,7,8,9], while the

latter ones are designed aiming at handling complexgenomes

[10,11,12].

One of the most intractable bottlenecks for practic

al assembly

of next - generation short reads is how toprocess rep

etitive

fragmentsfrom

complicated

genomes,especia

llyeukaryote

genomes.Intuitively, sequencing with

longer reads is a potential

solution, whileit becomes

impractical with

limit current of

sequencing technology. The paired-end (PE) sequencing can, to

someextent, compensate for read length [13]. Several assemblers

,

suchas SSAKE [9], SOAPdenovo [11], AbySS [12], Velvet

[14,15], exploit PE sequencing information to reduce gaps from

assembled contigs. Another

big challenge for the assembly of

short reads is the intensive computational time req

uirement. To

decreasethe tim

e costof the assembly

procedure,thread

parallelization is

implemented

ina couple

of graph-based

assemblers [11,12].

At our lastenumeration, 24 academic

denovo

genome

assemblers,

eachpossessin

g itsown range of applica

tion, are

developedfor short reads datasets from

different sequencing

PLoS ONE | www.plosone.org

1

March2011 | Volume 6 | Issu

e 3 | e17915

A Practical Compariso

n of De Novo Genome Assembly

Software Tools for Next-Generation Sequencing

Technologies

Wenyu Zhang, JiajiaChen, Yang Yang, Yifei Tang, Jing Shang, Bairong Shen*

Center for Systems Biology, Soochow University, Suzhou, Jiangsu, China

Abst ract

The advent of next-generation sequencing technologies is accompanied with

the development of many whole-genome

sequence assembly methods and software, especially for denovo fragment assembly. Due to the poor knowledge about the

applicability

and performanceof these software tools, choosing a befitti

ng assembler becomes a tough task. Here, we

provide the information of adaptivityfor each

program, then above all, compare the performanceof eight distin

ct tools

against eight groups of simulated datasets from

Solexa sequencing platform. Considering the computational time,

maximum random access memory (RAM) occu

pancy, assembly accuracy and integrity, our study indicate that strin

g-based

assemblers,overlap-layout-co

nsensus (OLC) assemblers are well-suited for very short reads and longer reads of small

genomes respectively. For large datasets of more than hundred millio

ns of short reads, De Bruijn graph-based assemblers

would be more appropriate. In terms of software implementation, string-based assemblers are superior to graph-based

ones, of whichSOAPdenovo is complex for the creation of configuration file. Our compariso

n study will assist researchers in

selecting a well-su

ited assembler and offer essential information for the improvement of existing assemblers or the

developing of novel assemblers.

Citat ion: Zhang W, Chen J, Yang Y, Tang Y, Shang J, et al. (2011) A Practical Compariso

n of De Novo Genome Assembly Software Tools for Next-Generation

Sequencing Technologies. PLoS ONE 6(3): e17915. doi:10.1371/journal.pone.0017915

Editor: I. King Jordan, Georgia Institute of Technology, United States of America

Received November 20, 2010; Accept ed February 15, 2011; Published March14, 2011

Copyright:2011 Zhang et al. This is an open-acce

ss article distri

buted under the terms of the Creative Commons Attribution License, which

permits

unrestricted use, distri

bution, and reproduction in any medium, provided the original author and source

are credited.

Funding: This work was supported by grants from the National Nature ScienceFoundation of China (31040020) and the National 973 Programs of China

(2007CB947002, 2010CB945600). The funders had no role in study design, data collection and analysis,

decision to publish

, or preparation of the manuscript.

Compet ing Interests: The authors have declared that no competing interestsexist.

* E-mail: bairong.shen@suda.edu.cn

Introduct ion

Inrece

nt years,the next-generation sequencing (or deep

sequencing) technologies

have beenevolving rapidly, with

the

potential toaccel

eratebiological and

biomedical research

dramatically

[1]. However,the downstre

amanalysis

of short

reads datasetsafter

sequencing is a tough task; one of the biggest

challenges for the analysis of high throughput sequencing reads is

the whole genome assembly. DNA fragment assem

bly has a long

history since the emergence of the firstgeneration of sequencing

technologies

[2,3]. The assembly proced

ure becomes especia

lly

difficult when

tackling short and high throughput reads with

different erro

r profiles[4]. According to the existen

ce of reference

information, theassembly proced

ure can beclassified asrefe

rence-

guidegenome assembly and denovo

genome assembly, of which

the

former is relatively

toilless with

the aid of reference genome or

proteome information while the laterin morechallen

ging. Herein,

we focus on the comparison and evaluation of tools for de novo

assembly of genome sequence.

The genome assemblersgenerally take a file

of short sequence

reads and a fileof quality-value as the input. Since the quality-

value filefor the high throughput short reads is usually

highly

memory-intensive, onlya

fewassemblers

, forexample,

SHARCGS [5], and ALLPATHS-LG[6] adopt it in

the

posterior assembly

process. For the sake of computational

memory saving and convenience of data inquiry, high-through-

put short reads data is always initially formattedto specif

ic data

structure. Curren

tly, existing data stru

cture for this usage can be

predominantlyclassifi

edinto two categories

: string-based

model

and graph-basedmodel.

We therefore call the corres

ponding

assemblers as string-based

and graph-based. String-based

assem-

blers, implemented with

Greedy-extension algorithm, are mainly

reported for the assembly of small genomes [5,7,8,9], while the

latter ones are designedaiming at handling complex

genomes

[10,11,12].

One of the most intractable bottlenecks for practic

al assembly

of next - generation short reads is how to processrepetiti

ve

fragmentsfrom

complicated genomes,

especiallyeukaryote

genomes.Intuitively, sequencing with

longer reads is a potential

solution, whileit beco

mesimpractic

al withlimit current of

sequencing technology. The paired-end (PE) sequencing can, to

someextent, compensate for read length [13]. Several assemblers

,

suchas SSAKE [9], SOAPdenovo [11], AbySS [12], Velvet

[14,15], exploit PE sequencing information to reduce gaps from

assembled contigs. Another big challenge for the assembly of

short reads is the intensive computational time requirem

ent. To

decrease

the time costof the assembly

procedure,

thread

parallelization is implem

entedin

a coupleof graph-based

assemblers [11,12].

At our lastenumeration, 24 academic

denovo

genome

assemblers,each

possessing its

own range of application, are

developed for short reads datasets fromdiffer

ent sequencing

PLoS ONE | www.plosone.org

1

March2011 | Volume 6 | Issu

e 3 | e17915

platforms in the last few years. These assembly tools with

corresponding websites and references are listed in Table S1. We

classify and list these assemblersaccording to their data structure

models in Figure 1. In the present study, eight short reads

assemblers, representing four various assembly strategies, were

benchmarked against two types of simulated short reads datasets

derived from four different genomes. Our objective is to gain the

assemblers performance information about computational time

and memory cost, assembly accuracy, completeness and size

distribution of assembled contigs when each assembler is applied

to handle datasets with different data size, then to provide

essential information for researchers in choosing suitable tools

and for computational biologists to develop novel assemblers.

The result indicates that each assembly strategy has its own

range of applicability while PE readsand longer readsare indeed

with the capability to increase the quality of assembled contigs to

some extent, and parallel computing is of great potential in short

reads assembly, with which the computational time is notably

reduced.

Results

At present, mainly three distinct strategies are applied in short

reads assembly. Among them, Greedy-extension is the implemen-

tation of string-based method, while DeBruijn graph and overlap-

layout-consensus (OLC) are two different graph-based approaches.

Each assembly tool issuitable for dataset from specific sequencing

platform.

For each short reads assembly procedure, less computational

time and memory cost is our expectation. The computational

time of the assembly process is determined by both the dataset

complexity and the assembly strategy. The information about

running times, maximum memory occupancies for different

assemblers applied to different datasets is illustrated in Figures 2

and 3. For string-based assembler, the time and memory cost is

approximately proportionate to dataset size, although it is also

affected by the complexity of dataset. Among them, SSAK E

runs in rather less time than other peer assemblers, but the

RAM usage increases dramatically with augmentation of dataset

size. QSRA [7] is developed upon the original VCAK E

algorithm, which indeed reduces the computational time, at

the cost of RAM occupation. SHARCGS runs in comparable

speed as QSRA, however it is highly memory-intensive, even

unable to handle E.coli short reads dataset with our computer

power used in this study. Edena is a typically graph-based

assembly tool, which has two running modes: strict and nonstrict

modes [16]. For the strict mode, fewer but more accurate

contigs are generated, while nonstrict mode acts on the contrary.

Compared with string-based tools, Edena is superior in terms of

time and RAM utilization. Velvet and SOAPdenovo typify

another graph-based method. Similar to Edena, they implement

Figure 1. Overview of de novo short reads assemblers. Programs developed from year of 2005 to 2010 are classified according to the assemblystrategies. Currently, there are mainly four sorts of assemblers, while the other ones are denoted as Different box symbols areutilized to distinguish assemblers that for short reads from different platforms.doi:10.1371/journal.pone.0017915.g001

Comparison of De Novo Genome Assembly Tools

PLoS ONE | www.plosone.org 2 March 2011 | Volume 6 | Issue 3 | e17915

9/7/2013

21

platforms in the last few years. These assembly tools with

corresponding websites and references are listed in Table S1. We

classify and list these assemblersaccording to their data structure

models in Figure 1. In the present study, eight short reads

assemblers, representing four various assembly strategies, were

benchmarked against two types of simulated short reads datasets

derived from four different genomes. Our objective is to gain the

assemblers performance information about computational time

and memory cost, assembly accuracy, completeness and size

distribution of assembled contigs when each assembler is applied

to handle datasets with different data size, then to provide

essential information for researchers in choosing suitable tools

and for computational biologists to develop novel assemblers.

The result indicates that each assembly strategy has its own

range of applicability while PE readsand longer readsare indeed

with the capability to increase the quality of assembled contigs to

some extent, and parallel computing is of great potential in short

reads assembly, with which the computational time is notably

reduced.

Results

At present, mainly three distinct strategies are applied in short

reads assembly. Among them, Greedy-extension is the implemen-

tation of string-based method, while DeBruijn graph and overlap-

layout-consensus (OLC) are two different graph-based approaches.

Each assembly tool issuitable for dataset from specific sequencing

platform.

For each short reads assembly procedure, less computational

time and memory cost is our expectation. The computational

time of the assembly process is determined by both the dataset

complexity and the assembly strategy. The information about

running times, maximum memory occupancies for different

assemblers applied to different datasets is illustrated in Figures 2

and 3. For string-based assembler, the time and memory cost is

approximately proportionate to dataset size, although it is also

affected by the complexity of dataset. Among them, SSAK E

runs in rather less time than other peer assemblers, but the

RAM usage increases dramatically with augmentation of dataset

size. QSRA [7] is developed upon the original VCAK E

algorithm, which indeed reduces the computational time, at

the cost of RAM occupation. SHARCGS runs in comparable

speed as QSRA, however it is highly memory-intensive, even

unable to handle E.coli short reads dataset with our computer

power used in this study. Edena is a typically graph-based

assembly tool, which has two running modes: strict and nonstrict

modes [16]. For the strict mode, fewer but more accurate

contigs are generated, while nonstrict mode acts on the contrary.

Compared with string-based tools, Edena is superior in terms of

time and RAM utilization. Velvet and SOAPdenovo typify

another graph-based method. Similar to Edena, they implement

Figure 1. Overview of de novo short reads assemblers. Programs developed from year of 2005 to 2010 are classified according to the assemblystrategies. Currently, there are mainly four sorts of assemblers, while the other ones are denoted as Different box symbols areutilized to distinguish assemblers that for short reads from different platforms.doi:10.1371/journal.pone.0017915.g001

Comparison of De Novo Genome Assembly Tools

PLoS ONE | www.plosone.org 2 March 2011 | Volume 6 | Issue 3 | e17915

Melhor performance para shot-reads muito pequenos

Melhor performance para datasets muito grandes

9/7/2013

22

Maiores reads = menosbolhas = menos contigs

Testes estatísticos confirmar o qual completo e

contíguo está o genoma/transcriptoma

Teste N50

N50 = bons contigs ou scaffolds = genoma bom

N50 = muitos contigs ou scaffolds pequenos = genoma mal

sequenciado = genoma mal montado

% gaps % bases não sequenciadas (Ns) Ns = genoma bom

% cobertura % do genoma sequenciado. % = genoma bom

A cobertura é baseada em dados citológicos (90%-95% é consi. bom)

9/7/2013

23

Yandell e Ence 2012

N50 próximo mediana do

gene = ~50% dos genes

estarão em scaffolds ou

contigs únicos