Efficient Assembly of Large Genomes

Shaun Jackman

Efficient Assembly
of Large Genomes

1. Introduction
2. ABySS 2.0
3. Tigmint
4. UniqTag
5. ORCA
6. Organellar genomes of white spruce
7. Mitochondrial genome of Sitka spruce
8. Genome assembly of western redcedar
9. Conclusion

Short Read Genome Assembly

ABySS 1.0

• de Bruijn graph assembler
• Stored k-mers in a hash table
• Distributed the hash table over many machines
• Used MPI to aggregate sufficient memory
• Assembles large genomes

ABySS 1.0

Challenges

• High memory usage
• Interprocess communication is slow
• Intermachine communication is really slow

Solution

• A memory-efficient data structure
  reduces memory usage
• Fitting entire graph in a single machine
  eliminates intermachine communication
• OpenMP rather than MPI
  eliminates interprocess communication

ABySS 2.0

Spruce genome assemblies

* Time of unitig assembly without scaffolding

ABySS 2.0 Conclusions

• ABySS 2.0 reduces memory usage by 10 fold
  from 418 GB for ABySS 1.0
  to 34 GB for ABySS 2.0
  for a human genome assembly
• High-throughput short-read sequencing
  combined with large molecule scaffolding
  such as 10X Genomics, BioNano, Hi-C
  permits cost effective assembly of large genomes

Linked Reads

Tools for Linked Reads

https://github.com/johandahlberg/awesome-10x-genomics

Contigs and scaffolds
come to an end due to…

• repeats
• sequencing gaps
• structural variation
• misassemblies

Misassemblies limit contiguity

particularly for highly contiguous assemblies.

Most scaffolding tools do not correct misassemblies.

Method

• Assemble human HG004 with PE, MP, and linked reads
• Scaffolding with ARCS improved NGA50 from 3 to 8 Mbp
• Tigmint reduced misassemblies by 216 (27% reduction)
• Tigmint + ARCS improved NGA50 over five-fold to 16 Mbp

Note: Supernova used only linked reads, others PE+MP+LR.

Time and Memory

Tigmint Conclusions

Scaffolding after correcting with Tigmint yields an assembly both more correct and more contiguous.

Linked reads permit cost-effective assembly of large genomes using high-throughput sequencing.

Western Redcedar Methods

• Trim adapters with Trimadap and NxTrim
• Count k-mers with ntCard
• Estimate genome size GenomeScope
• Assemble PE and MP reads with ABySS 2.0
• Correct assembly errors
  with Chromium reads using Tigmint
• Scaffold with Chromium reads using ARCS
• Assess genome completeness using BUSCO

Western Redcedar Results

• 12.5 Gbp genome size estimated by flow cytometry
  (Hizume et al. 2001 https://doi.org/d89svf)
• 9.8 Gbp genome size estimated by GenomeScope
• 7.95 Gbp assembled in scaffolds 1 kbp or larger
• 2.31 Mbp scaffold N50
• 1.71 Mbp scaffold NG50 (with G=10 Gbp)
• Tigmint improved NG50 by 14% over ARCS alone
• BUSCO 60.4% of core single-copy genes present
  53.9% complete, 6.5% fragmented, 39.6% missing

Efficient Assembly
of Large Genomes

1. Introduction
2. ABySS 2.0
3. Tigmint
4. UniqTag
5. ORCA
6. Organellar genomes of white spruce
7. Mitochondrial genome of Sitka spruce
8. Genome assembly of western redcedar
9. Conclusion

Thesis Committee

Research Supervisors

Committee Members

University Examiners

Physlr

Physical Maps of Linked Reads

Our Dance Card

👫 Human (3 Gbp)

🌲 Western redcedar (12 Gbp)

🌲 Sitka spruce (20 Gbp)

fin

Supplemental Slides

Publications

• Four first-author (or joint) papers
• One paper each year from 2015 through 2018
• Collaborated on 32 papers since 2009
• 28 papers with at least 10 citations
• One first-author manuscript in review (ORCA)
• One first-author manuscript in preparation
  (Sitka spruce mitochondrion)
• ABySS has been cited over 2,700 times!

First-author Publications

• Tigmint: correcting assembly errors using linked reads from large molecules
  SD Jackman, L Coombe, J Chu, RL Warren, BP Vandervalk, …
  BMC Bioinformatics 2018
• ABySS 2.0: resource-efficient assembly of large genomes using a Bloom filter
  SD Jackman^*, BP Vandervalk^*, H Mohamadi, J Chu, S Yeo, SA Hammond, …
  Genome Research 2017
• Organellar genomes of white spruce (Picea glauca): assembly and annotation
  SD Jackman, RL Warren, EA Gibb, BP Vandervalk, H Mohamadi, J Chu, …
  Genome Biology and Evolution 2015
• UniqTag: content-derived unique and stable identifiers for gene annotation
  SD Jackman, J Bohlmann, I Birol
  PLOS ONE 2015

Selected Publications

• Assembly of the complete Sitka spruce chloroplast… L Coombe, RL Warren, SD Jackman, C Yang, BP Vandervalk, …, I Birol
  PloS one 2016
• Spaced seed data structures for de novo assembly
  I Birol, J Chu, H Mohamadi, SD Jackman, K Raghavan, …, RL Warren
  International journal of genomics 2015
• Konnector v2.0: pseudo-long reads from PE sequencing
  BP Vandervalk, C Yang, Z Xue, K Raghavan, J Chu, H Mohamadi, SD Jackman, …, I Birol
  BMC medical genomics 2015
• Sealer: a scalable gap-closing application…
  D Paulino, RL Warren, BP Vandervalk, A Raymond, SD Jackman, I Birol
  BMC Bioinformatics 2015
• On the representation of de Bruijn graphs
  R Chikhi, A Limasset, SD Jackman, JT Simpson, P Medvedev
  Journal of Computational Biology 2015
• Improved white spruce (Picea glauca) genome…
  RL Warren, CI Keeling, MMS Yuen, A Raymond, GA Taylor, …, J Bohlmann
  The Plant Journal 2015
• Assembling the 20Gb white spruce genome…
  I Birol, A Raymond, SD Jackman, S Pleasance, R Coope, …, SJM Jones
  Bioinformatics 2013