De novo Genome Assembly

Transcription

1 De novo Genome Assembly A/Prof Torsten Seemann Winter School in Mathematical & Computational Biology - Brisbane, AU - 3 July 2017

2 Introduction

3 The human genome has 47 pieces MT (or XY)

4 The shortest piece is 48,000,000 bp Total haploid size 3,200,000,000 bp (3.2 Gbp) Total diploid size 6,400,000,000 bp (6.4 Gbp) = 6,400,000,000 bp

5 In an ideal world... AGTCTAGGATTCGCTATAG ATTCAGGCTCTGATATATT TCGCGGCATTAGCTAGAGA TCTCGAGATTCGTCCCAGT CTAGGATTCGCTAT AAGTCTAAGATTC... Human DNA isequencer 46 chromosomal and 1 mitochondrial sequences

6 The real world ( for now ) Genome Short fragments Sequencing Reads are stored in FASTQ files

7 Read lengths bp bp 5,000-40,000+ bp 1, ,000+ bp

8

9 Assemble Compare

10 Genome assembly (the red pill)

11 De novo genome assembly Reconstruct the original genome from the sequence reads only Millions of short sequences (reads) A few long sequences (contigs) Ideally, one sequence per replicon.

12 De novo genome assembly From scratch

13 An example

14 A small genome I ll return them tomorrow! Friends, Romans, countrymen, lend me your ears;

15 Shakespearomics Reads ds, Romans, count ns, countrymen, le Friends, Rom send me your ears; crymen, lend me Whoops! I dropped them.

16 Shakespearomics Reads ds, Romans, count ns, countrymen, le Friends, Rom send me your ears; crymen, lend me I am good with words. Overlaps Friends, Rom ds, Romans, count ns, countrymen, le crymen, lend me send me your ears;

17 Shakespearomics Reads ds, Romans, count ns, countrymen, le Friends, Rom send me your ears; crymen, lend me We have reached a consensus! Overlaps Friends, Rom ds, Romans, count ns, countrymen, le crymen, lend me send me your ears; Majority consensus Friends, Romans, countrymen, lend me your ears; (1 contig)

18 Overlap - Layout - Consensus Amplified DNA Shear DNA Sequenced reads Overlaps Layout Consensus Contigs

19 Assembly graphs (not Excel bar charts)

20 Overlap graph

21 Another example is this

22 Overlaps find

23 The graph one can simplify Not, look is fully self contained within the other longer read, and can be removed.

24 Do the graph traverse Size matters not. Look at me. Judge me by my size, do you? Size matters not. Look at me. Judge me by my soze, do you? 2 supporting reads 1 supporting reads

25 So far, so good.

26

27 Why is it so hard?

28 What makes a jigsaw puzzle hard? Lots of pieces No box Frayed pieces Missing pieces Repetitive regions :., Dirty pieces Multiple copies No corners (circular genomes)

29 What makes genome assembly hard 1. Many pieces (read length is very short compared to the genome) Size of the human genome = 3.2 x 109 bp (3,200,000,000) Typical short read length = 102 bp (100) A puzzle with millions to billions of pieces Storing in RAM is a challenge succinct data structure

30 What makes genome assembly hard 2. Lots of overlaps Finding overlaps means examining every pair of reads Comparisons = N (N-1)/2 ~ N2 Lots of smart tricks to reduce this close to ~N

31 What makes genome assembly hard 3. Lots of sky (short repeats) ATATATATATATATATATATATATATATATATATATATATATATATATATATATATATATATA TATATATA TATATATA TATATATA TATATATA TATATATA TATATATA TATATATA TATATATA TATATATA TATATATA TATATATA How could we possibly assemble this segment of DNA? All the reads are the same!

32 What makes genome assembly hard 4. Dirty pieces (sequencing errors) Read 1: GGAACCTTTGGCCCTGT Read 2: GGCGCTGTCCATTTTAGAAACC What counts as overlapping? Minimum overlap length Minimum DNA indentity

33 What makes genome assembly hard 5. Multiple copies (long repeats) Gene Copy 1 Gene Copy 2 Our old nemesis the REPEAT!

34 Repeats

35 What is a repeat? A segment of DNA that occurs more than once in the genome

36 Major classes of repeats in the human genome Repeat Class Arrangement Coverage (Hg) Length (bp) Satellite (micro, mini) Tandem 3% SINE Interspersed 15% Transposable elements Interspersed 12% 200-5k LINE Interspersed 21% 500-8k rdna Tandem 0.01% 2k-43k Segmental Duplications Tandem or Interspersed 0.2% 1k-100k

37 Repeats Repeat copy 1 Repeat copy 2 1 locus 4 contigs Collapsed repeat consensus

38 Repeats cause graph ambiguity Repeat copy 1 Repeat copy 2 1 locus Overlap graph

39 Repeats are hubs in the graph

40 Long reads can span repeats Repeat copy 1 Repeat copy 2 long reads 1 contig

41 Draft vs Finished genomes (bacteria) 250 bp - Illumina - $200 10,000 bp - Pacbio - $2000

42 The two laws of repeats 1. It is impossible to resolve repeats of length L unless you have reads longer than L. 2. It is impossible to resolve repeats of length L unless you have reads longer than L.

43 How much data do we need?

44 Coverage vs Depth Coverage (aka Breadth) Depth Fraction of the genome sequenced by at least one read Average number of reads that cover any given region Intuitively more reads should increase coverage and depth

45 Example: Read length = ⅛ Genome Length Coverage = ⅜ Depth = ⅜ Genome Reads

46 Example: Read length = ⅛ Genome Length Coverage = 4.5/8 Depth = 5/8 Genome Reads Newly Covered Regions = 1.5 Reads

47 Example: Read length = ⅛ Genome Length Coverage = 5.2/8 Depth = 7/8 Genome Reads Newly Covered Regions = 0.7 Reads

48 Example: Read length = ⅛ Genome Length Coverage = 6.7/8 Depth = 10/8 Genome Reads Newly Covered Regions = 1.5 Reads

49 Depth is easy to calculate Depth = N L / G Depth = Y / G N = Number of reads L = Length of a read G = Genome length Y = Sequence yield (N x L)

50 Example: Coverage of the Tarantula genome The size estimate of the tarantula genome based on k-mer analysis is 6 Gb and we sequenced at 40x coverage from a single female A. geniculata Sangaard et al Nature Comm (5) p1-11 Depth = N L / G 40 = N x 100 / (6 Billion) N = 40 * 6 Billion / 100 N = 2.4 Billion

51 Coverage and depth are related Approximate formula for coverage assuming random reads Coverage = 1 - e-depth

52 Much more sequencing needed in reality Sequencing is not random GC and AT rich regions are under represented Other chemistry quirks More depth needed for: sequencing errors polymorphisms Pacbio Illumina (both)

53 Assessing assemblies

54 Contiguity Completeness Correctness

55 Contiguity Desire Fewer contigs Longer contigs Metrics Number of contigs Average contig length Median contig length Maximum contig length N50, NG50, D50

56 Contiguity: the N50 statistic

57 Completeness : Total size Proportion of the original genome represented by the assembly Can be between 0 and 1 Proportion of estimated genome size but estimates are not perfect

58 Completeness: core genes Proportion of coding sequences can be estimated based on known core genes thought to be present in a wide variety of organisms. Assumes that the proportion of assembled genes is equal to the proportion of assembled core genes. In the past this was done with a tool called CEGMA There is a new tool for this called BUSCO Number of Core Genes in Assembly Number of Core Genes in Database

59 Correctness Proportion of the assembly that is free from mistakes Errors include Mis-joins Repeat compressions Unnecessary duplications Indels / SNPs caused by assembler

60 Correctness: check for self consistency Align all the reads back to the contigs Look for inconsistencies Assembly Align Original Reads Mapped Read

61 Assemble ALL the things

62 Why assemble genomes Produce a reference for new species Genomic variation can be studied by comparing against the reference A wide variety of molecular biological tools become available or more effective Coding sequences can be studied in the context of their related non-coding (eg regulatory) sequences High level genome structure (number, arrangement of genes and repeats) can be studied

63 Vast majority of genomes remain unsequenced Eukaryotes Bacteria & Archaea Estimated number of species ~9M Estimated number of species Millions to Billions Whole genome sequences (including incomplete drafts) ~ 3000 Genomic sequences ~ 600,000 Every individual has a different genome Some diseases such as cancer give rise to massive genome level variation

64

65 Not just genomes Transcriptomes One contig for every isoform Do not expect uniform coverage Meta-genomes Mixture of different organisms Host, bacteria, virus, fungi all at once All different depths Meta-transcriptomes Combination of above!

66 Conclusions

67 Take home points De novo assembly is the process of reconstructing a long sequence from many short ones Represented as a mathematical overlap graph Assembly is very challenging ( impossible ) because sequencing bias under represents certain regions Reads are short relative to genome size Repeats create tangled hubs in the assembly graph Sequencing errors cause detours and bubbles in the assembly graph

68 Acknowledgments Nick Hamilton Ira Cooke Ann-Marie Patch Katia Nones Simon Gladman Mark Ragan Anna Syme

69 Contact

70 The End