Genome assembly Henrik Lantz - BILS/SciLife/Uppsala University De novo genome project workflow • • • • • • • • • Extracting DNA (and RNA) - as much DNA as possible! Choosing best sequence technology for the project Sequencing Quality assessment and other pre-assembly investigations Assembly Assembly validation Assembly comparisons Repeat masking? Annotation Genome assembly - things to think about • Genome specifics - Size of genome, number of chromosomes, repeat content, heterozygosity • Which assembly programs can run on “my” genome? • What kind of data do these programs need? Genome assembly - things to think about Genome assembly - things to think about • Genome specifics - Size of genome, number of chromosomes, repeat content, heterozygosity • Which assembly programs can run on “my” genome? • What kind of data do these programs need? • How much data do I need? Will I have enough coverage? Do I need to subsample? • Are there closely related organisms that already have had their genome sequenced? • Do I need additional data for post-assembly? Genome assembly programs • • • • • • • • • • • Abyss Allpaths-LG CABOG (a.k.a. Celera) HGAP Masurca Mira Newbler SGA SoapDeNovo Spades Velvet Genome assembly programs Name Algorithm Data Abyss De Bruijn Illumina Allpaths-lg De Bruijn Illumina/PacBio CABOG (Celera) OLC All HGAP OLC PacBio Masurca De Bruijn/OLC All Mira “OLC” All Newbler OLC 454/Illumina/Torrent SGA String Illumina SoapDeNovo De Bruijn Illumina Spades De Bruijn Illumina Velvet De Bruijn Illumina OLC vs. de Bruijn de Bruijn de Bruijn Sequence Assembly via De Bruijn Graphs From Martin & Wang, Nat. Rev. Genet. 2011 From Martin & Wang, Nat. Rev. Genet. 2011 From Martin & Wang, Nat. Rev. Genet. 2011 De Bruijn • Pros: Computationally efficient, can work with large coverage short read datasets • Cons: Sensitive to sequence errors, connection between assembly and read is lost, does not work so well with longer reads OLC • Pros: Utilizes longer reads well • Cons: Time consuming, high memory requirements Assemblathon 2 • Uses 454, Illumina, and PacBio for three large eukaryote genomes: a bird, a fish, and a snake • Bird - Illumina 14 libraries, 454, PacBio • Fish - Illumina, 8 libraries • Snake - Illumina, 4 libraries • Teams take the data, perform assemblies with whatever tools they wish, and then submit their results => teams are evaluated more than individual programs! GigaScience 2013, 2:10 Assemblathon 2 Assemblathon 2 - Bird vs. Snake Assemblathon 2 - Bird CEGMA Assemblathon 2 - Validation measures GAGE-B • Uses Illumina (HiSeq and MiSeq) data for a number of bacteria • One team runs all programs => assembly programs are compared, not teams • Reference high quality assemblies are available => errors/misassemblies can be quantified Comparison of N50 contig size (in kilobases) on the y-axis, versus depth of coverage on the x-axis, for the eight assemblers used in this study. Magoc T et al. Bioinformatics 2013;29:1718-1725 © The Author 2013. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com GAGE-B GAGE-B statistics Genome assembly programs - pros and cons • • • • • • • • • • Abyss Allpaths-LG CABOG (a.k.a. Celera) Masurca Mira Newbler SGA SoapDeNovo Spades Velvet Allpaths-LG • Pros: Produces contigs and scaffolds with high N50 values, can use PacBio data for scaffolding, can run on large genomes with high coverage • Cons: Only accepts Illumina data, needs very specific libraries to work at all (180 bp + 3 kbp), needs very high coverage (100x), takes a long time to run and requires a lot of memory Assemblathon 2 - Bird vs. Snake Assemblathon 2 - Bird Masurca • Pros: Can accept any type of data, is a true hybrid assembler, usable for very large genomes, produces top results in comparison of assembly statistics • Cons: Takes a long time to run, unstable(?) Comparison of N50 contig size (in kilobases) on the y-axis, versus depth of coverage on the x-axis, for the eight assemblers used in this study. Magoc T et al. Bioinformatics 2013;29:1718-1725 © The Author 2013. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com GAGE-B statistics MIRA • Pros: Can accept any type of data, is a true hybrid assembler, produces good assemblies for smaller genomes, excellent documentation • Cons: Only useful for smaller genomes (bacteria, fungi), can not use high coverage data (prefers max 50x), takes a long time to run, no scaffolding Comparison of N50 contig size (in kilobases) on the y-axis, versus depth of coverage on the x-axis, for the eight assemblers used in this study. Magoc T et al. Bioinformatics 2013;29:1718-1725 © The Author 2013. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com GAGE-B statistics SoapDeNovo • Pros: Usable on large genomes, easy to use and runs fairly quickly, can use high coverage data • Cons: Only accepts Illumina data, medium results in assembly statistic comparisons Assemblathon 2 - Bird vs. Snake Assemblathon 2 - Bird GAGE-B statistics Spades • Pros: Designed to work with amplified data, performs very well in GAGE-B with MiSeq data • Cons: Only accepts Illumina data, only for smaller genomes Comparison of N50 contig size (in kilobases) on the y-axis, versus depth of coverage on the x-axis, for the eight assemblers used in this study. Magoc T et al. Bioinformatics 2013;29:1718-1725 © The Author 2013. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com GAGE-B statistics Some recommendations • • • • • • Large eukaryote genome, Illumina data: Allpaths-LG (needs specific libraries), SoapDeNovo, SGA, Masurca Large eukaryote genome, additional longer reads: Masurca, Newbler, CABOG Small eukaryote or prokaryote genome, Illumina data: Spades, Masurca, SoapDeNovo, Abyss, Velvet Small eukaryote or prokaryote genome, mixed data: MIRA, Masurca, Newbler Need to run in parallel: Abyss Amplified data (Single Cell Genomics): Spades Assemblathon 2 recommendations • • • • • • Based on the findings of Assemblathon 2, we make a few broad suggestions to someone looking to perform a de novo assembly of a large eukaryotic genome: 1. Don’t trust the results of a single assembly. If possible, generate several assemblies (with different assemblers and/or different assembler parameters). Some of the best assemblies entered for Assemblathon 2 were the evaluation assemblies rather than the competition entries. 2. Do not place too much faith in a single metric. It is unlikely that we would have considered SGA to have produced the highest ranked snake assembly if we had only considered a single metric. 3. Potentially choose an assembler that excels in the area you are interested in (e.g., coverage, continuity, or number of error free bases). 4. If you are interested in generating a genome assembly for the purpose of genic analysis (e.g., training a gene finder, studying codon usage bias, looking for intronspecific motifs), then it may not be necessary to be concerned by low N50/NG50 values or by a small assembly size. 5. Assess the levels of heterozygosity in your target genome before you assemble (or sequence) it and set your expectations accordingly. Post assembly considerations • External scaffolders: SSPACE (commercial), SGA (see Hunt et al. in Genome Biology 2014:15, R42). • Gap Closers (use with caution!): IMAGE, PILON, GapCloser • Error correction: Nesoni, PILON • Assembly validation is extremely important! Abyss • Pros: Only assembler that can run in parallel on different nodes => does not need a single huge memory node, fast, can run on large genomes with a high coverage • Cons: Only accepts Illumina data, does not excel in any statistics Comparison of N50 contig size (in kilobases) on the y-axis, versus depth of coverage on the x-axis, for the eight assemblers used in this study. Magoc T et al. Bioinformatics 2013;29:1718-1725 © The Author 2013. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com Assemblathon 2 - Bird vs. Snake CABOG (Celera) • Pros: Can accept any type of data, is a true hybrid assembler, output can easily be analyzed in the assembly validation toolkit AMOSvalidate, usable for large genomes • Cons: Does not perform so well for any statistic in comparisons Comparison of N50 contig size (in kilobases) on the y-axis, versus depth of coverage on the x-axis, for the eight assemblers used in this study. Magoc T et al. Bioinformatics 2013;29:1718-1725 © The Author 2013. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com GAGE-B statistics Newbler • Pros: Easy to run, works very well on 454 and Ion Torrent data, can use many types of data, usable for larger genomes, produces competitive assemblies if longer reads are available • Cons: Requires longer reads to perform well Assemblathon 2 - Bird vs. Snake Assemblathon 2 - Bird SGA • Pros: Usable on large genomes, memory-efficient • Cons: Only accepts Illumina data, does not perform well in comparisons of assembly statistics GAGE-B statistics Assemblathon 2 - Bird Velvet • Pros: Easy to use, runs quickly • Cons: Only accepts Illumina data, only for smaller genomes, does not excel in any assembly statistic comparison Comparison of N50 contig size (in kilobases) on the y-axis, versus depth of coverage on the x-axis, for the eight assemblers used in this study. Magoc T et al. Bioinformatics 2013;29:1718-1725 © The Author 2013. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com GAGE-B statistics