Torsten Seemann - de novo genome assembly

2,381 views

Published on

De novo assembly is the process of reconstructing a genome's DNA sequence using only a set of much shorter error‐prone sequences (reads) sampled from the genome. It is the "original" genomics‐based bioinformatics problem, because it is all we can do when we don't have any related reference genome sequences, with the exemplar being the original human genome project. This presentation will discuss the principles of and approaches to de novo assembly of data, and practical issues like computational and memory requirements, limitations of de novo assembly, terminology, file formats, available software, and an example run‐through of an assembly using the Velvet software if time permits.

First presented at the 2014 Winter School in Mathematical and Computational Biology http://bioinformatics.org.au/ws14/program/

Published in: Science

Torsten Seemann - de novo genome assembly

  1. 1. De novo genome assembly Dr Torsten Seemann IMB Winter School - Brisbane – Mon 7 July 2014
  2. 2. Introduction
  3. 3. Ideal world I would not need to give this talk! AGTCTAGGATTCGCTA CAGATTCAGGCTCTGA AGCTAGATCGCTATGC TATGATCTAGATCTCG AGATTCGTATAAGTCT AGGATTCGCTATAGAT TCAGGCTCTGATATAT Human DNA iSequencer™ 46 complete haplotype chromosome sequences
  4. 4. Real world •  Can’t sequence full-length native DNA –  no instrument exists (yet) •  But we can sequence short fragments – 100 at a time (Sanger) – 100,000 at a time (Roche 454) – 1,000,000 at a time (PGM) – 10,000,000 at a time (Proton, MiSeq) – 100,000,000 at a time (HiSeq)
  5. 5. De novo assembly The process of reconstructing the original DNA sequence from the fragment reads alone. •  Instinctively like a jigsaw puzzle – Find reads which “fit together” (overlap) – Could be missing pieces (sequencing bias) – Some pieces will be dirty (sequencing errors)
  6. 6. An example
  7. 7. A small “genome” Friends, Romans, countrymen, lend me your ears; I’ll return them tomorrow!
  8. 8. Shakespearomics •  Reads ds, Romans, count ns, countrymen, le Friends, Rom send me your ears; crymen, lend me Oops! I dropped them.
  9. 9. Shakespearomics •  Reads ds, Romans, count ns, countrymen, le Friends, Rom send me your ears; crymen, lend me •  Overlaps Friends, Rom ds, Romans, count ns, countrymen, le crymen, lend me send me your ears; I’m good with words.
  10. 10. Shakespearomics •  Reads ds, Romans, count ns, countrymen, le Friends, Rom send me your ears; crymen, lend me •  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; We have a consensus!
  11. 11. So far, so good.
  12. 12. The awful truth “Genome assembly is impossible.” A/Prof. Mihai Pop World leader in de novo assembly research. He wears glasses so he must be smart :-P
  13. 13. Methods
  14. 14. Approaches •  greedy assembly •  overlap :: layout :: consensus •  de Bruijn graphs •  string graphs •  seed and extend … all essentially doing the same thing, but taking different short cuts.
  15. 15. Assembly recipe •  Find all overlaps between reads – hmm, sounds like a lot of work… •  Build a graph – a picture of read connections •  Simplify the graph – sequencing errors will mess it up a lot •  Traverse the graph – trace a sensible path to produce a consensus
  16. 16. Clean graph
  17. 17. Find read overlaps •  If we have N reads of length L – we have to do ½N(N-1) ~ O(N²) comparisons – each comparison is an ~ O(L²) alignment – use special tricks/heuristics to reduce these! •  What counts as “overlapping” ? – minimum overlap length eg. 20bp – minimum %identity across overlap eg. 95% – choice depends on L and expected error rate
  18. 18. What we are up against!
  19. 19. What ruins the graph? •  Read errors – introduce false edges and nodes •  Non-haploid organisms – heterozygosity causes lots of detours •  Repeats – if longer than read length – causes nodes to be shared, locality confusion
  20. 20. Graph simplification •  Squash small bubbles – collapse small errors (or minor heterozygosity) •  Remove spurs – short “dead end” hairs on the graph •  Join unambiguously connected nodes – reliable stretches of unique DNA
  21. 21. Graph traversal •  For each unconnected graph –  at least one per replicon in original sample •  Find a path which visits each node once –  Hamiltonian path/cycle is NP-hard (this is bad) –  solution will be a set of paths which terminate at decision points •  Form a consensus sequences from paths –  use all the overlap alignments –  each of these collapsed paths is a contig
  22. 22. Contigs Contiguous, unambiguous stretches of assembled DNA sequence •  Contigs ends correspond to – Real ends (for linear DNA molecules) – Dead ends (missing sequence) – Decision points (forks in the road)
  23. 23. Repeats
  24. 24. What is a repeat? A segment of DNA which occurs more than once in the genome sequence •  Very common – Transposons (self replicating genes) – Satellites (repetitive adjacent patterns) – Gene duplications (paralogs)
  25. 25. Effect on assembly The repeated element is collapsed into a single contig
  26. 26. Repeat mis-assembly a b c a c b a b c d I II III I II III a bc d b c a b dc e f I II III IV I III II IV a d be c f a collapsed tandem excision rearrangement
  27. 27. The law of repeats •  It is impossible to resolve repeats of length S unless you have reads longer than S. •  It is impossible to resolve repeats of length S unless you have reads longer than S.
  28. 28. Scaffolding
  29. 29. Beyond contigs Contig sizes are limited by: •  the length of repeats in your genome – can’t change this! •  the length (or “span”) of the reads – wait for new technology – use “tricks” with existing technology
  30. 30. Paired reads •  DNA fragment (200-800 bp) ============================== •  Single end -------->=====================! •  Paired end (up to 800 bp span) ----->==================<-----! •  Mate pair (up to 20 kbp span) ---->========/+/=========<----!
  31. 31. Scaffolding •  Paired-end reads – known sequences at either end – roughly known distance between ends – unknown sequence between ends •  Most ends will occur in same contig – if our contigs are longer than pair distance •  Some ends will be in different contigs – evidence that these contigs are linked!
  32. 32. Contigs to scaffolds Contigs Paired-end read Scaffold Gap Gap
  33. 33. Assessment
  34. 34. Assessing assemblies •  We desire – Total length similar to genome size – Fewer, larger contigs – No mistakes (mis-assemblies) •  Metrics – No generally useful objective measure – Longest contig, total bp, N50, …
  35. 35. The “N50” The length of that contig from which 50% of the bases are in it and shorter contigs •  Imagine we got 7 contigs with lengths: – 1,1,3,5,8,12,20 •  Total – 1+1+3+5+8+12+20 = 50 •  N50 is the “halfway sum” = 25 – 1+1+3+5+8+12 = 30 (≥ 25) so N50 is 12
  36. 36. N50 concerns •  Optimizing for N50 –  encourages mis-assemblies! •  An aggressive assembler may over-join: – 1,1,3,5,8,12,20 (previous) – 1,1,3,5,20,20 (now) – 1+1+3+5+20+20 = 50 (unchanged) •  N50 is the “halfway sum” (still 25) – 1+1+3+5+20= 30 (≥ 25) so N50 is 20 (was 12)
  37. 37. Validation •  Self consistency – Align read back to contigs – Check for errors or discordant pairs •  Second opinion – Use two complementary sequencing methods – Target troublesome areas for PCR – Use a genome wide “optical map”
  38. 38. How can I play?
  39. 39. Considerations •  Size of genome – bacteria, eukaryote, meta-genome •  Hardware – phone, laptop, desktop, server, cloud – RAM is more limiting than CPU •  Operating system – Linux, Mac, Windows •  Software budget –  commercial, free, open-source
  40. 40. Recommendations •  SPAdes – Unix command-line (Mac, Linux) •  VAGUE (Velvet) – Unix GUI (Mac, Linux) •  CLC Genomics Workbench – Java GUI (Windows, Mac, Linux) – Commercial product
  41. 41. Online tutorial •  The GVL – Genomics Virtual Laboratory – http://genome.edu.au •  Protocols – Microbial de novo assembly for Illumina data – Written by Simon Gladman (VBC/LSCC) – https://genome.edu.au/wiki/Protocols
  42. 42. Contact •  Email – torsten.seemann@monash.edu •  Blog – TheGenomeFactory.blogspot.com •  Web – vicbioinformatics.com – vlsci.org.au/lscc – genome.edu.au Torst! ~10!

×