Microbial Genomics and Bioinformatics: BM405 (2015)

1. Microbial Genomics and Bioinformatics BM405 1.Introduction Leighton Pritchard1,2,3 1 Information and Computational Sciences, 2 Centre for Human and Animal Pathogens in the Environment, 3 Dundee Eﬀector Consortium, The James Hutton Institute, Invergowrie, Dundee, Scotland, DD2 5DA

2. Acceptable Use Policy Recording of this talk, taking photos, discussing the content using email, Twitter, blogs, etc. is permitted (and encouraged), providing distraction to others is minimised. These slides will be made available on SlideShare. These slides, and supporting material including exercises, are available at https://github.com/widdowquinn/Teaching- Strathclyde-BM405

3. Table of Contents Introduction A personal view Erwinia carotovora subsp. atroseptica Dickeya spp., Campylobacter spp., and Escherichia coli So what’s changed? High Throughput Sequencing Three revolutions, four dominant technologies Benchmarking Nanopore How fast is sequence data increasing? Sequence Data Formats FASTQ SAM/BAM/CRAM Repositories Assembly Overlap-Layout-Consensus de Bruijn graph assembly Read Mapping Short-Read Sequence Alignment The Assembly What you get back Comparative Genomics Computational Comparative Genomics Bulk Genome Properties Nucleotide Frequency/Genome Size Whole Genome Alignment An Introduction to Pairwise Genome Alignment Average Nucleotide Identity Whole Genome Alignment in Practice Ordering Draft Genomes By Alignment Chromosome painting Nosocomial P.aeruginosa acquisition Genome Features What are genome features? Prokaryotic CDS Prediction Assessing Prediction Methods Prokaryotic Annotation Pipelines Genome-Scale Functional Annotation Functional Annotation A visit to the doctor Statistics of genome-scale prediction Building to Metabolism Reconstructing metabolism Equivalent Genome Features What makes genome features equivalent? Homology, Orthology, Paralogy Who let the -logues out? What’s so important about orthologues? Evaluating orthologue prediction Using orthologue predictions Core and Pan-genomes Conclusions Things I Didn’t Get To Conclusions

4. The impacta a Loman and Pallen (2015) Nat. Rev. Micro. doi:10.1038/nrmicro3565 Genome sequencing and bioinformatics have transformed our understanding of prokaryotic biology: • function • evolution • interactions • community structure • real-time monitoring and diagnostics • as a platform for synthetic biology It now takes much longer to analyse than generate data

5. The endpoints • 2003: Erwinia carotovora subsp. atroseptica • 2015: Dickeya spp., Campylobacter spp., and Escherichia coli

7. 2003: E. carotovora subsp. atroseptica • £250k collaboration between SCRI, University of Cambridge, WT Sanger Institute • Single isolate: E. carotovora subsp. atroseptica SCRI1043 • The ﬁrst sequenced enterobacterial plant pathogen (32 authors!) 1 • All repeats and gaps bridged and sequenced directly • Result: a single, complete, high-quality 5Mbp circular chromosome at 10.2X coverage: 106,500 reads 1 Bell et al. (2004) Proc. Natl. Acad. Sci. USA 101: 30:11105-11110. doi:10.1073/pnas.0402424101

8. 2003: E. carotovora subsp. atroseptica A genome sequence is a starting point. . . • Manual annotation by the Sanger Pathogen Sequencing Unit • Literature searches and comparisons • Six people, for six months ≈ three person-years • Genes: BLAST, GLIMMER, ORPHEUS • Functional domains: PFAM, SIGNALP, TMHMM • Metabolism: KEGG • ncRNA: RFAM

9. 2003: E. carotovora subsp. atroseptica Working (Eca_Sanger_annotation.gbk) and published (NC_004547.gbk) annotation ﬁles are in the data directory

10. 2003: E. carotovora subsp. atroseptica Compared against all 142 available bacterial genomes2 2 data/Pba directory in the accompanying GitHub repository

12. 2013: Dickeya spp. Sequenced and annotated 25 new isolates of Dickeya • 25 Dickeya isolates, at least six species • Multiple sequencing methods: 454, Illumina (SE, PE) • Minor publications (6, 8 authors)3,4 • Results: 12-237 fragments containing 4.2-5.1Mbp, at 6-84X coverage, 170k-4m reads • Automated annotation: RAST with manual corrections 3 Pritchard et al. (2013) Genome Ann. 1 (4) doi:10.1128/genomeA.00087-12 4 Pritchard et al. (2013) Genome Ann. 1 (6) doi:10.1128/genomeA.00978-13

13. 2013: Dickeya spp. Within-genus comparisons: large-scale synteny and rearrangement Within-species comparisons: e.g. indels, HGT

14. 2013: Dickeya spp. Within-genus comparisons: whole genome-based species delineation5 5 van der Wolf et al. (2014) Int. J. Syst. Evol. Micr. 64:768-774 doi:10.1099/ijs.0.052944-0

15. 2013: Dickeya spp. Within-genus comparisons: diﬀerences in metabolism

16. 2014: E. coli Sequenced and annotated ≈ 190 isolates of E. coli All bacteria environmental, sampled from lysimeters • Illumina paired-end sequencing. Total cost of sequencing 190 bacteria: ≈£11k • Automated annotation: PROKKA

17. 2014: E. coli Sequencing output variable - even though same preps, “same” bacteria, similar sources. • Results: 5-3000 contigs (median ≈ 125); 9kbp-7.1Mbp (median ≈ 5Mbp); 170k-4m reads

18. 2014: E. coli Genome sequencing enables within-species classiﬁcation Brunei20070942_contigs Muenster20063091_contigs Senftenberg20070885_contigs Lys142_contigs Lys175_contigs Lys130_contigs Lys170_contigs Lys126_contigs Lys167_contigs Lys176_contigs Lys169_contigs Lys50_contigs X5038_contigs Lys131_contigs Lys171_contigs Lys111_contigs Lys107_contigs Lys114_contigs Lys16_contigs Lys22_contigs Lys65_contigs Lys56_contigs Lys113_contigs Lys109_contigs Lys77_contigs Lys102_contigs Lys100_contigs Lys92_contigs Lys94_contigs Lys80_contigs Lys64_contigs Lys82_contigs AW3_contigs X5008_contigs AW4_contigs AW1_contigs Lys118_contigs Lys138_contigs Lys121_contigs Lys122_contigs Lys177_contigs Lys155_contigs Lys165_contigs Lys163_contigs Lys160_contigs Lys161_contigs Lys172_contigs Lys144_contigs Lys135_contigs Lys146_contigs Lys123_contigs Lys124_contigs Lys150_contigs Lys140_contigs Lys157_contigs Lys173_contigs Lys156_contigs Lys158_contigs Lys159_contigs Lys162_contigs Lys5_contigs X5084_contigs X5042_contigs Lys110_contigs Lys136_contigs Lys54_contigs Lys1_contigs Lys6_contigs Lys112_contigs X5012_contigs Lys30_contigs Lys25_contigs Lys43_contigs Lys37_contigs Lys40_contigs Lys151_contigs Lys31_contigs Lys27_contigs Lys42_contigs Lys51_contigs Lys33_contigs Lys46_contigs Lys38_contigs Lys89_contigs Lys23_contigs Lys115_contigs Lys108_contigs Lys104_contigs DSM10973_contigs Lys125_contigs Lys105_contigs Lys17_contigs Lys128_contigs Lys66_contigs Lys73_contigs Lys15_contigs Lys91_contigs DSM8698_contigs DSM8695_contigs Lys74_contigs Lys61_contigs Lys9_contigs Lys153_contigs Lys84_contigs Lys93_contigs Lys72_contigs Lys62_contigs Lys21_contigs Lys59_contigs Lys63_contigs Lys83_contigs Lys19_contigs Lys4_contigs AW13_contigs Lys45_contigs Lys28_contigs Lys53_contigs Lys52_contigs Lys34_contigs Lys36_contigs Lys24_contigs Lys35_contigs Lys68_contigs Lys106_contigs Lys88_contigs Lys97_contigs Lys76_contigs Lys134_contigs Lys58_contigs Lys71_contigs Lys81_contigs Lys129_contigs Lys120_contigs Lys145_contigs Lys137_contigs Lys127_contigs Lys152_contigs Lys101_contigs Lys98_contigs Lys70_contigs Lys133_contigs Lys47_contigs Lys75_contigs Lys48_contigs Lys148_contigs Lys139_contigs Lys141_contigs Lys164_contigs Lys149_contigs Lys147_contigs Lys60_contigs Lys79_contigs Lys168_contigs Lys18_contigs Lys87_contigs Lys96_contigs Lys7_contigs Lys154_contigs Lys117_contigs Lys119_contigs Lys178_contigs Lys116_contigs Lys86_contigs Lys90_contigs Lys41_contigs Lys13_contigs Lys85_contigs X5002_contigs Lys12_contigs Lys39_contigs Lys14_contigs Lys55_contigs Lys29_contigs Lys99_contigs X5035_contigs Lys8_contigs Lys3_contigs X5034_contigs X5088_contigs Lys20_contigs Lys78_contigs Lys11_contigs Brunei20070942_contigs Muenster20063091_contigs Senftenberg20070885_contigs Lys142_contigs Lys175_contigs Lys130_contigs Lys170_contigs Lys126_contigs Lys167_contigs Lys176_contigs Lys169_contigs Lys50_contigs 5038_contigs Lys131_contigs Lys171_contigs Lys111_contigs Lys107_contigs Lys114_contigs Lys16_contigs Lys22_contigs Lys65_contigs Lys56_contigs Lys113_contigs Lys109_contigs Lys77_contigs Lys102_contigs Lys100_contigs Lys92_contigs Lys94_contigs Lys80_contigs Lys64_contigs Lys82_contigs AW3_contigs 5008_contigs AW4_contigs AW1_contigs Lys118_contigs Lys138_contigs Lys121_contigs Lys122_contigs Lys177_contigs Lys155_contigs Lys165_contigs Lys163_contigs Lys160_contigs Lys161_contigs Lys172_contigs Lys144_contigs Lys135_contigs Lys146_contigs Lys123_contigs Lys124_contigs Lys150_contigs Lys140_contigs Lys157_contigs Lys173_contigs Lys156_contigs Lys158_contigs Lys159_contigs Lys162_contigs Lys5_contigs 5084_contigs 5042_contigs Lys110_contigs Lys136_contigs Lys54_contigs Lys1_contigs Lys6_contigs Lys112_contigs 5012_contigs Lys30_contigs Lys25_contigs Lys43_contigs Lys37_contigs Lys40_contigs Lys151_contigs Lys31_contigs Lys27_contigs Lys42_contigs Lys51_contigs Lys33_contigs Lys46_contigs Lys38_contigs Lys89_contigs Lys23_contigs Lys115_contigs Lys108_contigs Lys104_contigs DSM10973_contigs Lys125_contigs Lys105_contigs Lys17_contigs Lys128_contigs Lys66_contigs Lys73_contigs Lys15_contigs Lys91_contigs DSM8698_contigs DSM8695_contigs Lys74_contigs Lys61_contigs Lys9_contigs Lys153_contigs Lys84_contigs Lys93_contigs Lys72_contigs Lys62_contigs Lys21_contigs Lys59_contigs Lys63_contigs Lys83_contigs Lys19_contigs Lys4_contigs AW13_contigs Lys45_contigs Lys28_contigs Lys53_contigs Lys52_contigs Lys34_contigs Lys36_contigs Lys24_contigs Lys35_contigs Lys68_contigs Lys106_contigs Lys88_contigs Lys97_contigs Lys76_contigs Lys134_contigs Lys58_contigs Lys71_contigs Lys81_contigs Lys129_contigs Lys120_contigs Lys145_contigs Lys137_contigs Lys127_contigs Lys152_contigs Lys101_contigs Lys98_contigs Lys70_contigs Lys133_contigs Lys47_contigs Lys75_contigs Lys48_contigs Lys148_contigs Lys139_contigs Lys141_contigs Lys164_contigs Lys149_contigs Lys147_contigs Lys60_contigs Lys79_contigs Lys168_contigs Lys18_contigs Lys87_contigs Lys96_contigs Lys7_contigs Lys154_contigs Lys117_contigs Lys119_contigs Lys178_contigs Lys116_contigs Lys86_contigs Lys90_contigs Lys41_contigs Lys13_contigs Lys85_contigs 5002_contigs Lys12_contigs Lys39_contigs Lys14_contigs Lys55_contigs Lys29_contigs Lys99_contigs 5035_contigs Lys8_contigs Lys3_contigs 5034_contigs 5088_contigs Lys20_contigs Lys78_contigs Lys11_contigs ANIm 0.9 0.92 0.94 0.96 0.98 Value 0100020003000400050006000 Color Key and Histogram Count A B1 B2 C D E F U X

19. 2014: Campylobacter spp. Sequenced ≈ 1034 isolates of Campylobacter Clinical, animal, food-associated isolates • Illumina paired-end sequencing. Total cost of sequencing >1000 bacteria: ≈£60k • Automated annotation: PRODIGAL

20. 2014: Campylobacter spp. • Identiﬁed 15554 gene families from genecalls. • To calculate, took 23 days on institute cluster (4e12 pairwise protein comparisons!).

22. So what’s changed? • Cost: £250k per genome, to £60 per genome. Now cheaper to sequence a genome than to analyse it! • Location: sequencing centre, to benchtop • Data: volume has increased massively - what you get back from machines, and what’s out there to work with More data is better, but also more challenging. • Speed: typical sequencing run time can be less than a day • Software: more software to do more things (but not always better. . .) • New kinds of experiment: genomes, exomes, variant calling, methylated sequences, . . . • New kinds of application: diagnostics, epidemic tracking, metagenomics, . . .

23. So what’s changed? Having a single genome is useful, but having thousands really helps comparative genomics: combining genomic data, evolutionary and comparative biology • Transfer functional understanding of model systems (e.g. E. coli) to non-model organisms • Genomic differences may underpin phenotypic (host range, virulence, physiological) differences • Genome comparisons aid identification of functional elements on the genome • Studying genomics changes reveals evolutionary processes and constraints

25. Revolutions One and Twoa a Loman and Pallen (2015) Nat. Rev. Micro. doi:10.1038/nrmicro3565 Revolution One: whole-genome shotgun • First bacterial genomes: Haemophilus inﬂuenzae (1995); E. coli, Bacillus subtilis (1997) • (Oh, and the human genome) Revolution Two: high-throughput sequencing • ”Next-generation” sequencing (now ”last-generation”). • 454 GS20 (2005), Illumina GAII (2007). • metagenomics; surveillance sequencing; SNP-based comparisons; transposon-sequencing for functional genomics; ChIP-seq; . . .

26. Not all HT sequencing is the same It’s all about the biology, but it all starts with the data. Sequencing technology (including library prep.) aﬀects your sequence data. • Roche/454 • Illumina • Ion Torrent • Paciﬁc Bioscience (PacBio)

27. The basic principle DNA source is fragmented, and the fragments are sequenced.

28. HTS: PE vs SE High-throughput sequencing (e.g. Illumina), reads may be single-end, or paired-end. Putting the jigsaw back together is sequence assembly.

29. Four different chemistriesa a Loman et al. (2012) Nat. Rev. Micro. 31:294-296 doi:10.1038/nbt.2522 Reads differ by technology, and may require different bioinformatic treatment. . . • Roche/454: Pyrosequencing (long reads, but expensive, and high homopolymer errors) (700-800bp, 0.7Gbp, 23h) • Illumina: Reversible terminator (cost-effective, massive throughput, but short read lengths) (2x150bp, 1.5Gbp, 27h) • Ion Torrent: Proton detection (short run times, good throughput, high homopolymers errors) (200bp, 1Gbp, 3h) • PacBio: Real-time sequencing (very long reads, high error rate, expensive) (3-15kbp, 3Gbp/day, 20min) . . . different error profiles, varying capability to assemble/determine variation

30. Costs of sequencinga a Miyamoto et al. (2014) BMC Genomics 15:699 doi:10.1186/1471-2164-15-699

32. Benchmarked performance Apply several sequencing technologies to the same sample(s). Benchmark comparisons inform appropriate choice of sequencing technology6,7,8,9,10,11,12 Progress in technologies is driving research very rapidly. Always look for most recent/relevant benchmarks. Bioinformatic methods also need to be benchmarked. 6 Miyamoto et al. (2014) BMC Genomics 15:699 doi:10.1186/1471-2164-15-699 7 Salipante et al. (2014) Appl. Environ. Micro. 80:7583-7591 doi:10.1128/AEM.02206-14 8 Frey et al. (2014) BMC Genomics 15:96 doi:10.1186/1471-2164-15-96 9 Koshimizu et al. (2013) PLoS One 8:e74167 doi:10.1371/journal.pone.0074167 10 Quail et al. (2012) BMC Genomics 13:341 doi:10.1186/1471-2164-13-341 11 Loman et al. (2012) Nat. Biotech. 30:434-439 doi:10.1038/nbt.2198 12 Lam et al. (2011) Nat. Biotech. 1 (6) doi:10.1038/nbt.2065

33. Benchmarking on Vibrioa a Miyamoto et al. (2014) BMC Genomics 15:699 doi:10.1186/1471-2164-15-699 • Sequenced Vibrio parahaemolyticus (2x chromosomes, closed reference genome) with four technologies • Chose an assembler for each tech, and assembled reads • Excess reads with Ion/MiSeq: used random subsets of reads to determine required coverage • Aligned assemblies (MUMmer) to known high-quality chromosome sequence, to measure error

34. Benchmarking on Vibrioa a Miyamoto et al. (2014) BMC Genomics 15:699 doi:10.1186/1471-2164-15-699

35. Benchmarking on Vibrioa a Miyamoto et al. (2014) BMC Genomics 15:699 doi:10.1186/1471-2164-15-699 De novo assembly and alignment against Vibrio parahaemolyticus (2x chromosomes)

36. Benchmarking on Vibrioa a Miyamoto et al. (2014) BMC Genomics 15:699 doi:10.1186/1471-2164-15-699 • More and longer reads do not always give the best assemblies: read depth, read distribution, error rate also matters • Optimal assemblies were obtained at around 60x-80x coverage, for Illumina and Ion. • Multiple rRNA regions are fragmented in short-read assemblies • PacBio generated single chromosome contigs • Assembly of multiple-chromosome bacteria is currently feasible Variability in published genomes as methods are not standard (e.g. sequencing technology, assembler, parameter settings and pre-processing). . .

38. Revolution Threea a Loman and Pallen (2015) Nat. Rev. Micro. doi:10.1038/nrmicro3565 Revolution Three: single-molecule long-read sequencing • Living through the revolution, now • PacBio (SMRT): large machine, expensive • Nanopore: portable device, inexpensive • Less mature, less accurate, improving rapidly

39. The future dominant sequencer? Oxford Nanopore. A sequencer the size of your hand. • Microﬂuidics, single-molecule sequencing; 11-70kbp reads • Reports current across pore (tiny electron microscope) as molecule moves through • $10/Mbp, 110Mbp per ﬂowcell13 13 Yaniv Erlich (2013) Future Continuous blog

40. Early dataa a Quick et al. (2014) GigaScience 3:22 doi:10.1111/1755-0998.12324 It’s a fast-moving area, and results are improving.

41. Developing tools Oxford Nanopore’s open beta went out without analysis tools. Tools (Poretools, poRe, etc.) were written/tested/validated by the user community14,15 14 Loman and Quinlan (2014) Bioinformatics doi:10.1093/bioinformatics/btu555 15 Watson et al. (2014) Bioinformatics doi:10.1093/bioinformatics/btu590

42. Recent applications • Amplicon sequencing (16S metagenomics) of bacteria and viruses 16 • Real-time viral diagnostics 17 • Scaﬀolding of a bacterial genome 18 • Complete de novo assembly of a bacterial genome 19 16 Kilianski et al. (2015) GigaScience doi:10.1186/s13742-015-0051-z 17 Greninger et al. (2015) Genome Med. doi:10.1186/s13073-015-0220-9 18 Karlsson et al. (2015) Sci. Reports doi:10.1038/srep11996 19 Loman et al. (2015) Nat. Meth. doi:10.1038/nmeth.3444

43. The three revolutionsa a Loman and Pallen (2015) Nat. Rev. Micro. doi:10.1038/nrmicro3565

45. Predicting the future is hard. . . “How many genomes will we have, and when?” Su et al. attempted to answer this20: 20 http://sulab.org/2013/06/sequenced-genomes-per-year/

46. After that, the flood. . . High-throughput sequencing methods have completely changed the landscape of microbiology (Nearly) complete, (mainly) accurate sequence data is now inexpensive (and cheaper than analysis) • GOLD (19/2/2014): 3,011 “finished” ; 9,891 “permanent draft” genomes • GOLD (10/11/2015): 7,657 “finished” ; 27,438 “permanent draft” genomes; 50,673 prokaryotes • NCBI WGS (19/2/2014): 17,023 microbial genomes • NCBI Genome (10/11/2015): 55,033 prokaryotic genomes

47. Pseudomonas In 2011, 25 isolate sequences21; in 2015, 2098 genomes: We’re going to need bigger bioinformatics. . . 21 Studholme (2011) Mol. Plant Pathol. doi:10.1111/j.1364-3703.2011.00713.x

48. Microbial Genomics and Bioinformatics BM405 2.Assembly Leighton Pritchard1,2,3 1 Information and Computational Sciences, 2 Centre for Human and Animal Pathogens in the Environment, 3 Dundee Eﬀector Consortium, The James Hutton Institute, Invergowrie, Dundee, Scotland, DD2 5DA

50. What do you get from sequencing Sequence reads. Usually lots of them. Size/number/errors depend on technology used. 22 22 Miyamoto et al. (2014) BMC Genomics 15:699 doi:10.1186/1471-2164-15-699

51. Sequence Read Data Formats Two common read data sequence formats: • FASTQ: Related to FASTA, a de facto standard for sequence reads • SAM/BAM: Sequence alignment/mapping format, two ﬂavours - uncompressed and compressed New formats are required to handle very large numbers of genomes • CRAM: Reference-based sequence compression You might also receive assembled genomes directly from a sequencing partner

53. FASTQa a Cock et al. (2009) Bioinformatics 38:1767-1771 doi:10.1093/nar/gkp1137 @HISEQ2500-09:168:HA424ADXX:2:1101:1404:2061 1:N:0:ATCTCTCTCACCAACT CGGTCTTGGGATAGATGGGTTGCAGGTTGCGGTAAAGCTCGGACTCCAGAGCGTCCAGGGTAGACTGGCTAATCTTCTGCTCTTTATCGATCATTATTTC + @@CBDDFFHHDFDHEGHIICGIFHHIIIIFHGGHIEHHIIIIGHGHIIIIIGGHHFFFFC@CBCCCDDBDCDDDDDDDDCCDDDD3@ABDDDDDEEEDE@ Files typically have .fq, .fastq extension. Four lines per sequence 1. Header: sequence identiﬁer and optional description, starts with “@” 2. Raw sequence ([ACGTN]) 3. Optional header, repeats line 1, starts with “+” 4. Quality scores, numbers encoded as ASCII Qphred = −10 log10 e, where e is the estimated probability that a base call is incorrect (like a pH).

54. Quality Control The quality of basecalls (error rate) varies between and along reads. (real data from our E.coli sequencing: good quality)

55. Quality Control Some datasets are better than others. Reads can be trimmed, or discarded. Including poor reads compromises assembly.

56. FASTQ encodinga a Cock et al. (2009) Bioinformatics 38:1767-1771 doi:10.1093/nar/gkp1137 More than one version of FASTQ, diﬀer by quality encoding Numbers converted to ASCII start at diﬀerent values

57. FASTQ encodinga a Cock et al. (2009) Bioinformatics 38:1767-1771 doi:10.1093/nar/gkp1137 Versions vary by sequencer and period. Most now settled on Sanger format (occasionally see historical data). Quality scores (Qphred ) oﬀset to lie in the given range: 1. Sanger: 33-126, used in SAM/BAM, and Illumina 1.8+ 2. Illumina 1.0-1.2: 59-126 3. Illumina 1.3-1.8: 64-126 Knowing where your data comes from, and the data format and version, is always important.

59. SAMa a https://github.com/samtools/hts-specs Intended to represent read alignments, also used for raw reads. Tab-delimited plain text. Headers (optional) start with “@”

60. BAMa /CRAMb a https://github.com/samtools/hts-specs b http://www.ebi.ac.uk/ena/software/cram-toolkit BAM is a compressed version of SAM. • BGZF compression. • Random access within compressed ﬁle, through indexing. CRAM format may come to dominate, especially in archives, as datasets get larger: • Reference-based compression.23 • Highly suited to compression and archiving of very large amounts of sequence data.24 23 Fritz et al. (2011) Genome Res. 21:734-740 doi:10.1101/gr.114819.110 24 Cochrane et al. (2012) GigaScience 1:2 doi:10.1186/2047-217X-1-2

62. Read repositories Repositories are centrally-maintained locations that keep sequence read data from multiple projects Submission to a repository is a requirement for publication. And the right thing to do! • ENA: The European Nucleotide Archive (http://www.ebi.ac.uk/ena), maintained by EBI/EMBL • SRA: The Short Read Archive (http://www.ncbi.nlm.nih.gov/sra), maintained in the US by NCBI

63. Sequence Assembly Once you have reads, you can assemble a genome. Two main approaches to read assembly: • Overlap-Layout-Consensus: Typically used with smaller sets of longer reads (e.g. 454, PacBio, Ion, Nanopore) • de Bruijn assembly: Typically used with many, shorter reads (e.g. Illumina), but also useful for longer reads See e.g. Leland Taylor’s thesis (http://gcat.davidson.edu/phast/docs/Thesis PHAST LelandTaylor.pdf), and PHAST (http://gcat.davidson.edu/phast/index.html).

65. Overlap-Layout-Consensus

66. Overlap-Layout-Consensus The oldest approach, originally used with smaller sets of fewer reads. Can be time consuming (all-vs-all comparisons), but oﬀset with graph-based OLC algorithms (e.g. SGA). Now more important again, with long-read data. • Celera Assembler25 • Newbler (the Roche/454 GS assembler)26 • String Graph Assembler27 25 http://wgs-assembler.sourceforge.net/ 26 http://www.454.com/products/analysis-software/ 27 Simpson and Durbin (2012) Genome Res. 22:549-556 doi:10.1101/gr.126953.111

68. de Bruijn graph assembly Used for short reads (e.g. Illumina): k-mer based graph (choice of k important):

69. de Bruijn graph assembly k-mer based genome and read graphs28 “True” edges = genome; “Error” edges = wrong assembly 28 Chaisson et al. (2009) Genome Res. 19:336-346 doi:10.1101/gr.079053.108

70. de Bruijn graph assembly All sequencing technologies have basecall errors. • The proportion of errors is approximately constant per read • Basecall errors lead to edge errors • The more reads you have, the more errors there are Increased coverage does not ensure increased accuracy29 29 Conway and Bromage (2011) Bioinformatics 27:479-486 doi:10.1093/bioinformatics/btq697

71. de Bruijn graph assembly Fast, and scales well to large datasets, as it never computes all-against-all overlaps. Sensitive to sequencing errors, but resolves short repeats (graph bulges and whirls). Notable tools: • Velvet30 • CLC Assembly Cell31 • Cortex32 30 Zerbino and Birney (2008) Genome Res. 18:821-829 doi:10.1101/gr.074492.107 31 http://www.clcbio.com/products/clc-assembly-cell/ 32 Iqbal et al. (2012) Nat. Genet. 44:226-232 doi:10.1038/ng.1028

72. “Coloured” de Bruijn graph assemblies Cortex33 allows for on-the-fly identification of complex variation, and genotyping, by tracking “coloured” edges in the graph. Colours ≈ different isolates/organisms (e.g. a reference) 33 Iqbal et al. (2012) Nat. Genet. 44:226-232 doi:10.1038/ng.1028

74. Why map reads?a a Trapnell et al. (2009) Nat. Biotech. 27:455-457 doi:10.1038/nbt0509-455 “Resequencing” an organism (sequencing a close relative, looking for SNPs/indels) RNA-seq, ChIP-seq, etc. - coverage ≈ expression/binding To see where reads map on an assembled genome • Is coverage even? (can indicate repeats) • Are there SNPs/indels? (heterogeneous population) • Assembly problems?

75. Short-Read Sequence Alignmenta a Trapnell et al. (2009) Nat. Biotech. 27:455-457 doi:10.1038/nbt0509-455 An embarrassment of tools (over 60 listed on Wikipedia) Main approaches: • Alignment: Smith-Waterman mathematically guaranteed to be the best alignment available (e.g. BFAST, MOSAIK); approximation to S-W (e.g. BLAST); ungapped or gapped alignment (e.g. MAQ, FAST, mrFAST, SOAP). Can be slow. • Burrows-Wheeler Transform: Makes reusable index of the genome (e.g. Bowtie, BWA), can be extended to consider sequence probability (e.g. BWA-PSSM). Can be very fast. Other tools may employ diﬀerent algorithms, some designed to be parallelised on GPUs/FPGAs (e.g. NextGenMap, XpressAlign)

76. Visualising Read Mapping Several tools available, e.g. Tablet (the best. . .)34 34 Milne et al. (2013) Brief. Bioinf. 14:193-202 doi:10.1093/bib/bbs012

78. In an ideal world Ideally, you would have one sequence per chromosome/plasmid. (and no errors): a closed/complete genome. PacBio, Sanger, manual closing, Nanopore(?)

79. More realistically. . . Typically, a number of assembled fragments (contigs or scaﬀolds) are returned in FASTA format: a draft, disordered genome. Around 250 contigs for a 5Mbp genome is usual with Illumina

80. Ordering contigs Contigs can be ordered correctly into scaﬀolds if paired-end reads span gaps, or long reads are available (typically done during assembly). Gaps are usually ﬁlled with Ns (length estimated)

81. Ordering contigs Contigs and scaﬀolds can also be reordered by alignment to a reference genome. • Mauve/progressiveMauve35 • MUMmer36 35 Darling et al. (2004) Genome Res. 14:1394-1403 doi:10.1101/gr.2289704 36 Kurtz et al. (2004) Genome Biol. 5:R12 doi:10.1186/gb-2004-5-2-r12

82. Where next?a a Lefebure et al. (2010) Genome Biol. Evol. 2:646-655 doi:10.1093/gbe/evq048

83. Microbial Genomics and Bioinformatics BM405 3.Whole Genome Comparisons Leighton Pritchard1,2,3 1 Information and Computational Sciences, 2 Centre for Human and Animal Pathogens in the Environment, 3 Dundee Eﬀector Consortium, The James Hutton Institute, Invergowrie, Dundee, Scotland, DD2 5DA

86. The Power of Comparative Genomics Massively enabled by high-throughput sequencing, and the availability of thousands of sequenced isolates. Computational comparisons more powerful and precise than experimental comparative genomics: the ultimate microbial typing solution Three broad areas/scales: • Comparison of bulk genome properties • Whole genome sequence comparisons • Comparison of features/functional components

88. Nucleotide frequency/genome size • Very easy to calculate from complete/draft genome • Can calculate for individual contigs/scaﬀolds/regions • Usually reported in GUI genome browsers Trivial to determine using, e.g. Python

89. Nucleotide frequency/genome size GC content and chromosome size can be characteristic See data/bacteria size for example iPython notebook exercise

90. Blobologya a Kumar and Blaxter et al. (2011) Symbiosis 3:119-126 doi:10.1007/s13199-012-0154-6 Sequencing samples may be contaminated or contain microbial symbionts. Expect more host than symbiont/contaminant DNA GC content and read coverage can be used to separate contigs, following assembly and mapping http://nematodes.org/bioinformatics/blobology/

91. k-mers • Nucleotides: [ACGT] • Dinucleotides: [AA|AC|AG|AT|CA|CC|. . .] (16 dimers) • Trinucleotides: [AAA|AAC|AAG|AAT|ACA|. . .] (64 trimers) • k-mers: 4k k-mers (see example in data/shiny)

92. k-mers GC content = point value; k-mer frequencies = vector (list) Diagnostic diﬀerences in k-mer frequency, and variability. The basis of several comparison tools E.coli Mycoplasma spp.

94. What to align, and why? To be useful, aligned genomes should: • derive from a sufficiently recent common ancestor, so homologous regions can be identified • derive from a sufficiently distant common ancestor, so that there are “interesting” differences to be identified • help to answer your biological question

95. How to align, and why? Naive sequence aligners (Needleman-Wunsch, Smith-Waterman) are not appropriate for genome alignment • Computationally expensive on large sequences • Cannot handle rearrangements Very many alternative alignment algorithms proposed • megaBLAST http://www.ncbi.nlm.nih.gov/blast/html/megablast.html • MUMmer http://mummer.sourceforge.net/ • BLAT http://genome.ucsc.edu/goldenPath/help/blatSpec.html • LASTZ http://www.bx.psu.edu/∼rsharris/lastz/ • LAGAN http://lagan.stanford.edu/lagan web/index.shtml • and many, many more. . . Example exercises in data/whole_genome_alignment.

96. megaBLAST Optimised for speed, over BLASTN37 • Genome-level searches • Queries on large sequence sets • Long alignments of very similar sequence Uses the greedy algorithm by Zhang et al.38, not BLAST algorithm. • Concatenates queries (“query packing”) to improve performance • Two modes: megaBLAST and discontinuous (dc-megablast) for divergent sequences BLASTN now uses the megaBLAST algorithm by default 37 http://www.ncbi.nlm.nih.gov/blast/Why.shtml 38 Zhang et al. (2000) J. Comp. Biol. 7:203-214 doi:10.1089/10665270050081478

97. BLAST vs megaBLAST megaBLAST is faster, but does it give the same biological results? megaBLAST (top) and BLAST (bottom) pairwise comparisons:

98. BLAST vs megaBLAST Filter out weak matches - not quite identical:

99. MUMmera a Kurtz et al. (2004) Genome. Biol. 5:R12 doi:10.1186/gb-2004-5-2-r12 Uses suffix trees for pattern matching: very fast even for large sequences • Finds maximal exact matches • Memory use depends only on the reference sequence size Suffix trees: (http://en.wikipedia.org/wiki/Suffix tree) • Can be built and searched in O(n) time • But useful algorithms are nontrivial

100. The MUMmer algorithma a Kurtz et al. (2004) Genome. Biol. 5:R12 doi:10.1186/gb-2004-5-2-r12 1. Identify a non-overlapping subset of maximal exact matches: often Maximal Unique Matches (MUMs) 2. Cluster into alignment anchors 3. Extend between anchors to produce the final alignment This is the basis of a very flexible suite of programs that align different kinds of sequence: mummer, nucmer, promer • nucleotide and (more sensitive) “conceptual protein” alignments • used for genome comparisons, assembly scaffolding, repeat detection, . . . • the basis of other aligners/assemblers (e.g. Mugsy, AMOS)

101. MUMmer vs megaBLAST MUMmer identiﬁes fewer weak matches megaBLAST (top) and MUMmer (bottom) pairwise comparisons:

102. MUMmer vs megaBLAST Filter out weak BLAST matches - not quite identical:

104. DNA-DNA hybridisationa a Morello-Mora and Amann (2001) FEMS Micro. Rev. 25:39-67 doi:10.1016/S0168-6445(00)00040-1 • “Gold Standard” for prokaryotic taxonomy, since 1960s. “70% identity ≈ same species.” • Denature DNA from two organisms. • Allow to anneal. Reassociation ≈ similarity, measured as ∆T of denaturation curves. Proxy for sequence similarity - replace with genome analysis39? 39 Chan et al (2012) BMC Microbiol. 12:302 doi:10.1186/1471-2180-12-302

105. Average Nucleotide Identity (ANIb)a a Goris et al. (2007) Int. J. Syst. Biol. 57:81-91 doi:10.1099/ijs.0.64483-0 1. Break genomes into 1020t fragments 2. ANIb: Mean % identity of all BLASTN matches with > 30% identity and > 70% fragment coverage. • DDH:ANIb linear • DDH:%ID linear • 70%ID ≈ 95%ANIb

106. Average Nucleotide Identity (ANIm)a a Richter and Rossello-Mora (2009) Proc. Natl. Acad. Sci. USA 106:19126-19131 doi:10.1073/pnas.0906412106 1. Align genomes (MUMmer) 2. ANIm: Mean % identity of all matches • DDH:ANIm linear • 70%ID ≈ 95%ANIb TETRA: tetranucleotide frequency-based classiﬁer introduced in same paper.

107. ANI/TETRA comparison All three methods applied to Anaplasma spp. ANIb: A_phagocytophilum_NC_021881 A_phagocytophilum_NC_021880 A_phgocytophilum_NC_021879 A_phagocytophilum_NC_007797 A_centrale_NC_013532 A_marginale_NC_004842 A_marginale_NC_012026 A_marginale_NC_022760 A_marginale_NC_022784 A_phagocytophilum_NC_021881 A_phagocytophilum_NC_021880 A_phgocytophilum_NC_021879 A_phagocytophilum_NC_007797 A_centrale_NC_013532 A_marginale_NC_004842 A_marginale_NC_012026 A_marginale_NC_022760 A_marginale_NC_022784 ANIb 0.9 0.94 0.98 Value 02040 Color Key and Histogram Count ANIm: A_phgocytophilum_NC_021879 A_phagocytophilum_NC_007797 A_phagocytophilum_NC_021880 A_phagocytophilum_NC_021881 A_centrale_NC_013532 A_marginale_NC_012026 A_marginale_NC_004842 A_marginale_NC_022760 A_marginale_NC_022784 A_phgocytophilum_NC_021879 A_phagocytophilum_NC_007797 A_phagocytophilum_NC_021880 A_phagocytophilum_NC_021881 A_centrale_NC_013532 A_marginale_NC_012026 A_marginale_NC_004842 A_marginale_NC_022760 A_marginale_NC_022784 ANIm 0.9 0.94 0.98 Value0102030 Color Key and Histogram Count TETRA: A_phagocytophilum_NC_021880 A_phgocytophilum_NC_021879 A_phagocytophilum_NC_007797 A_phagocytophilum_NC_021881 A_centrale_NC_013532 A_marginale_NC_022760 A_marginale_NC_022784 A_marginale_NC_012026 A_marginale_NC_004842 A_phagocytophilum_NC_021880 A_phgocytophilum_NC_021879 A_phagocytophilum_NC_007797 A_phagocytophilum_NC_021881 A_centrale_NC_013532 A_marginale_NC_022760 A_marginale_NC_022784 A_marginale_NC_012026 A_marginale_NC_004842 TETRA 0.9 0.94 0.98 Value 02040 Color Key and Histogram Count ANIb discards information, relative to ANIm: less sensitive ANIb/ANIm ≈ evolutionary history; TETRA ≈ bulk composition

108. ANI in practice Practical applications40 (note: no gene content used) 34 Dickeya isolates: species structure 180 E.coli isolates: subtyping Brunei20070942_contigs Muenster20063091_contigs Senftenberg20070885_contigs Lys142_contigs Lys175_contigs Lys130_contigs Lys170_contigs Lys126_contigs Lys167_contigs Lys176_contigs Lys169_contigs Lys50_contigs X5038_contigs Lys131_contigs Lys171_contigs Lys111_contigs Lys107_contigs Lys114_contigs Lys16_contigs Lys22_contigs Lys65_contigs Lys56_contigs Lys113_contigs Lys109_contigs Lys77_contigs Lys102_contigs Lys100_contigs Lys92_contigs Lys94_contigs Lys80_contigs Lys64_contigs Lys82_contigs AW3_contigs X5008_contigs AW4_contigs AW1_contigs Lys118_contigs Lys138_contigs Lys121_contigs Lys122_contigs Lys177_contigs Lys155_contigs Lys165_contigs Lys163_contigs Lys160_contigs Lys161_contigs Lys172_contigs Lys144_contigs Lys135_contigs Lys146_contigs Lys123_contigs Lys124_contigs Lys150_contigs Lys140_contigs Lys157_contigs Lys173_contigs Lys156_contigs Lys158_contigs Lys159_contigs Lys162_contigs Lys5_contigs X5084_contigs X5042_contigs Lys110_contigs Lys136_contigs Lys54_contigs Lys1_contigs Lys6_contigs Lys112_contigs X5012_contigs Lys30_contigs Lys25_contigs Lys43_contigs Lys37_contigs Lys40_contigs Lys151_contigs Lys31_contigs Lys27_contigs Lys42_contigs Lys51_contigs Lys33_contigs Lys46_contigs Lys38_contigs Lys89_contigs Lys23_contigs Lys115_contigs Lys108_contigs Lys104_contigs DSM10973_contigs Lys125_contigs Lys105_contigs Lys17_contigs Lys128_contigs Lys66_contigs Lys73_contigs Lys15_contigs Lys91_contigs DSM8698_contigs DSM8695_contigs Lys74_contigs Lys61_contigs Lys9_contigs Lys153_contigs Lys84_contigs Lys93_contigs Lys72_contigs Lys62_contigs Lys21_contigs Lys59_contigs Lys63_contigs Lys83_contigs Lys19_contigs Lys4_contigs AW13_contigs Lys45_contigs Lys28_contigs Lys53_contigs Lys52_contigs Lys34_contigs Lys36_contigs Lys24_contigs Lys35_contigs Lys68_contigs Lys106_contigs Lys88_contigs Lys97_contigs Lys76_contigs Lys134_contigs Lys58_contigs Lys71_contigs Lys81_contigs Lys129_contigs Lys120_contigs Lys145_contigs Lys137_contigs Lys127_contigs Lys152_contigs Lys101_contigs Lys98_contigs Lys70_contigs Lys133_contigs Lys47_contigs Lys75_contigs Lys48_contigs Lys148_contigs Lys139_contigs Lys141_contigs Lys164_contigs Lys149_contigs Lys147_contigs Lys60_contigs Lys79_contigs Lys168_contigs Lys18_contigs Lys87_contigs Lys96_contigs Lys7_contigs Lys154_contigs Lys117_contigs Lys119_contigs Lys178_contigs Lys116_contigs Lys86_contigs Lys90_contigs Lys41_contigs Lys13_contigs Lys85_contigs X5002_contigs Lys12_contigs Lys39_contigs Lys14_contigs Lys55_contigs Lys29_contigs Lys99_contigs X5035_contigs Lys8_contigs Lys3_contigs X5034_contigs X5088_contigs Lys20_contigs Lys78_contigs Lys11_contigs Brunei20070942_contigs Muenster20063091_contigs Senftenberg20070885_contigs Lys142_contigs Lys175_contigs Lys130_contigs Lys170_contigs Lys126_contigs Lys167_contigs Lys176_contigs Lys169_contigs Lys50_contigs 5038_contigs Lys131_contigs Lys171_contigs Lys111_contigs Lys107_contigs Lys114_contigs Lys16_contigs Lys22_contigs Lys65_contigs Lys56_contigs Lys113_contigs Lys109_contigs Lys77_contigs Lys102_contigs Lys100_contigs Lys92_contigs Lys94_contigs Lys80_contigs Lys64_contigs Lys82_contigs AW3_contigs 5008_contigs AW4_contigs AW1_contigs Lys118_contigs Lys138_contigs Lys121_contigs Lys122_contigs Lys177_contigs Lys155_contigs Lys165_contigs Lys163_contigs Lys160_contigs Lys161_contigs Lys172_contigs Lys144_contigs Lys135_contigs Lys146_contigs Lys123_contigs Lys124_contigs Lys150_contigs Lys140_contigs Lys157_contigs Lys173_contigs Lys156_contigs Lys158_contigs Lys159_contigs Lys162_contigs Lys5_contigs 5084_contigs 5042_contigs Lys110_contigs Lys136_contigs Lys54_contigs Lys1_contigs Lys6_contigs Lys112_contigs 5012_contigs Lys30_contigs Lys25_contigs Lys43_contigs Lys37_contigs Lys40_contigs Lys151_contigs Lys31_contigs Lys27_contigs Lys42_contigs Lys51_contigs Lys33_contigs Lys46_contigs Lys38_contigs Lys89_contigs Lys23_contigs Lys115_contigs Lys108_contigs Lys104_contigs DSM10973_contigs Lys125_contigs Lys105_contigs Lys17_contigs Lys128_contigs Lys66_contigs Lys73_contigs Lys15_contigs Lys91_contigs DSM8698_contigs DSM8695_contigs Lys74_contigs Lys61_contigs Lys9_contigs Lys153_contigs Lys84_contigs Lys93_contigs Lys72_contigs Lys62_contigs Lys21_contigs Lys59_contigs Lys63_contigs Lys83_contigs Lys19_contigs Lys4_contigs AW13_contigs Lys45_contigs Lys28_contigs Lys53_contigs Lys52_contigs Lys34_contigs Lys36_contigs Lys24_contigs Lys35_contigs Lys68_contigs Lys106_contigs Lys88_contigs Lys97_contigs Lys76_contigs Lys134_contigs Lys58_contigs Lys71_contigs Lys81_contigs Lys129_contigs Lys120_contigs Lys145_contigs Lys137_contigs Lys127_contigs Lys152_contigs Lys101_contigs Lys98_contigs Lys70_contigs Lys133_contigs Lys47_contigs Lys75_contigs Lys48_contigs Lys148_contigs Lys139_contigs Lys141_contigs Lys164_contigs Lys149_contigs Lys147_contigs Lys60_contigs Lys79_contigs Lys168_contigs Lys18_contigs Lys87_contigs Lys96_contigs Lys7_contigs Lys154_contigs Lys117_contigs Lys119_contigs Lys178_contigs Lys116_contigs Lys86_contigs Lys90_contigs Lys41_contigs Lys13_contigs Lys85_contigs 5002_contigs Lys12_contigs Lys39_contigs Lys14_contigs Lys55_contigs Lys29_contigs Lys99_contigs 5035_contigs Lys8_contigs Lys3_contigs 5034_contigs 5088_contigs Lys20_contigs Lys78_contigs Lys11_contigs ANIm 0.9 0.92 0.94 0.96 0.98 Value 0100020003000400050006000 Color Key and Histogram Count A B1 B2 C D E F U X 40 van der Wolf et al. (2014) Int. J. Syst. Evol. Micr. 64:768-774 doi:10.1099/ijs.0.052944-0

110. Collinearity and Synteny Genome rearrangements occur, but there can still be conservation of sequence similarity and ordering. • Two elements are collinear if they lie in the same linear sequence • Two elements are syntenous (or syntenic) if: • (orig.) they lie on the same chromosome • (mod.) there is conservation of blocks of order within the same chromosome Signs of evolutionary constraints, like sequence conservation or synteny, may indicate functional genome regions.

111. Pyrococcus spp.a a Zivanovic et al. (2002) Nuc. Acids Res. 30:1902-1910 doi:10.1093/nar/30.9.1902 Comparison of Pyrococcus genomes (P. horikoshii, P. abyssi, P. furiosus) shows chromosome-shuﬄing. Transposition a major cause of genomic disruption

112. Vibrio mimicus a a Hasan et al. (2010) Proc. Natl. Acad. Sci. USA 107:21134-21139 doi:10.1073/pnas.1013825107 Chromosome C-II carries genes associated with environmental adaptation; C-I carries virulence genes. C-II has undergone extensive rearrangement; C-I has not. Suggests modularity of genome organisation, as a mechanism for adaptation (HGT, two-speed genome).

113. Serratia symbiotica a a Burke and Moran (2011) Genome Biol. Evol. 3:195-208 doi:10.1093/gbe/evr002 S. symbiotica is a recently evolved symbiont of aphids Massive genomic decay is an adaptation to the new environment.

115. Multiple genome alignment is hard Can we not just align all our genomes, together? No. Because it’s really, really hard. Analogous to problems with multiple sequence alignment (three or more sequences). • Computationally extremely expensive (O(Ln), L=length of sequence, n=number of sequences) • NP-complete problem: no known eﬃcient way to ﬁnd a solution Heuristic (approximate) methods are used, most commonly: • Progressive alignment • Iterative alignment

116. Mauvea a Darling et al. (2004) Genome Res. 14:1394-1403 doi:10.1101/gr.2289704 Progressive alignment tool, with a GUI. Application to nine enterobacteria: rearrangement of homologous backbone. Alternatives include MLAGAN41 and MUMmer42 41 Brudno et al. (2003) Genome Res. 13:721-731 doi:10.1101/gr.926603 42 Kurtz et al. (2004) Genome Biol. 5:R12 doi:10.1186/gb-2004-5-2-r12

117. Mauve algorithma a Darling et al. (2004) Genome Res. 14:1394-1403 doi:10.1101/gr.2289704 1. Find local alignments (multi-MUMs) 2. Build guide tree from multi-MUMs 3. Select subset of multi-MUMs as anchors, and partition into Local Collinear Blocks (LCBs): consistently ordered subsets 4. Progressive alignment against guide tree

118. Reordering contigsa a Darling et al. (2004) Genome Res. 14:1394-1403 doi:10.1101/gr.2289704 Mauve also enables draft genome reordering. Once LCBs are identiﬁed, can apply Mauve Contig Mover to reorder contigs Example exercise in data/whole_genome_alignment

120. Chromosome paintinga a Yahara et al. (2013) Mol. Biol. Evol. 30:1454-1464 doi:10.1093/molbev/mst055 “Chromosome painting” infers recombination-derived ‘chunks’ Genome’s haplotype constructed in terms of recombination events from a ‘donor’ to a ‘recipient’ genome

121. Chromosome paintinga a Yahara et al. (2013) Mol. Biol. Evol. 30:1454-1464 doi:10.1093/molbev/mst055 Recombination events summarised in a coancestry matrix. H. pylori: most within geographical bounds, but asymmetrical donation from Amerind/East Asian to European isolates.

123. P.aeruginosa nosocomial acquisitiona a Quick et al. (2014) BMJ Open 4: e006278. doi:10.1136/bmjopen-2014-006278 Motivation Nosocomial water transmission of P.aeruginosa an urgent concern Setup Burns patients (30) screened for P.aeruginosa on admission Samples taken from patients and environment All P.aeruginosa isolates (141) WGS sequenced Outcome Clustering of isolates by room and outlet Three patient isolates identical to water isolates from same room Bioﬁlm from thermostatic mixer valve a possible source

124. P.aeruginosa nosocomial acquisitiona a Quick et al. (2014) BMJ Open 4: e006278. doi:10.1136/bmjopen-2014-006278

125. P.aeruginosa nosocomial acquisitiona a Quick et al. (2014) BMJ Open 4: e006278. doi:10.1136/bmjopen-2014-006278 Methods • Illumina MiSeq WGS of 141 isolates • Metagenomic sequencing of bioﬁlm • Simulated sequencing of 55 published P. aeruginosa • BWA mapping against PAO1 reference genome • SNPs called with SAMtools & VarScan • ML reconstruction with FastTree • De novo assembly with Velvet for MLST prediction Sequences and bioinformatic methods shared online: http://www.github.com/joshquick/snp calling scripts

128. P.aeruginosa nosocomial acquisitiona a Quick et al. (2014) BMJ Open 4: e006278. doi:10.1136/bmjopen-2014-006278 Strengths A P. aeruginosa source could be tracked by WGS Insights into transmission: water to patient a likely route Sensitivity - identiﬁes microevolution Limitations Small sample size: 5/30 patients infected, gave 55/141 isolates Not clear that causal inferences are general 300-day sampling, not real-time crisis analysis Good existing reference genome set for this bacterium Sequencing cost: ≈£8k; Staﬀ cost: ≈£15k

129. Microbial Genomics and Bioinformatics BM405 4.Genome Features Leighton Pritchard1,2,3 1 Information and Computational Sciences, 2 Centre for Human and Animal Pathogens in the Environment, 3 Dundee Eﬀector Consortium, The James Hutton Institute, Invergowrie, Dundee, Scotland, DD2 5DA

132. Genome Features • Genome features are annotated regions of the genome. • Typically represent functional elements. • May be simple (single region), or complex (subfeatures)

133. Why annotate genome features? • Almost all use of genomics depends on annotation: annotation quality is critical to downstream use of genomics in biology • Annotation is curation (a live, active process), not cataloguing • Automated annotation from curated data (public databases) is the only game in town, given the data quantities we generate • But you can’t propagate something that doesn’t exist: up to 30% of metabolic activity has no known gene associated with it43 • Biocurators can spend as much time “de-annotating” literature-based annotations as entering new data44 43 Chen and Vitkup (2007) Trends Biotech. doi:10.1016/j.tibtech.2007.06.001 44 Bairoch (2009) Nat. Preced. doi:10.1038/npre.2009.3092.1

134. Gene Features Gene features have signiﬁcant substructure, especially in eukaryotes. • 5‘ UTR • translation start • intron start/stop • exon start/stop • translation stop • translation terminator • 3‘ UTR

135. ncRNA Features • tRNA - transfer RNA • rRNA - ribosomal RNA • CRISPRs - bacterial/archaeal defence (used for genome editing) • many other classes

136. Regulatory/Repeat Features Regulatory sites • transcription start sites • RNA polymerase binding sites • Transcription Factor Binding Sites (TFBS) Repetitive regions and mobile elements • tandem repeats • (retro-)transposable elements • phage inclusions

137. Principles of feature prediction Two main approaches to feature prediction: • ab initio prediction - start from ﬁrst principles, using only the genome sequence: • Unsupervised methods - not trained on a dataset • Supervised methods - trained on a dataset • homology matches • alignment to features from related organisms (comparative genomics, annotation transfer) • from known gene products (e.g. proteins, ncRNA) • from transcripts/other intermediates (e.g. ESTs, cDNA, RNAseq) Dedicated tools available for many diﬀerent classes of feature.

139. Prokaryotic CDS Prediction Methods Using CDS prediction as an illustrative example for all feature prediction. Sequence conservation (evolutionary constraint; an unsupervised, a priori method) can be useful • Prokaryotes “easier” than eukaryotes for gene/CDS prediction • Less uncertainty in predictions (isoforms, gene structure) • Very gene-dense (over 90% of chromosome is coding sequence) • No intron-exon structure

140. Prokaryotic CDS Prediction Methods ORFs are plentiful: • Problem is: “which possible ORF contains the true gene, and which start site is correct?” • Still not a solved problem

141. Finding Open Reading Frames The simplest approach: ﬁnd ORFs (sequence between two consecutive in-frame stop codons) • ORF ﬁnding is naive, does not consider: • Start codon • Promoter/RBS motifs • Wider context (e.g. overlapping genes) Dedicated tools, e.g. Glimmer, Prodigal, RAST, GeneMarkS usually better.

142. Two ab initio CDS Prediction Tools • Glimmer45 • Interpolated Markov models • Can be trained on “gold standard” datasets • Prodigal46 • Log-likelihood model based on GC frame plots, followed by dynamic programming • Can be trained on “gold standard” datasets Applying these to an example bacterial chromosome. . . 45 Delcher et al. (2007) Bioinformatics 23:673-679 doi:10.1093/bioinformatics/btm009 46 Hyatt et al. (2010) BMC Bioinf. 11:119 doi:10.1186/1471-2105-11-119

143. Comparing predictions in Artemisa a Carver et al. (2012) Bioinformatics 28:464-469 doi:10.1093/bioinformatics/btr703 Not every ORF (green) is predicted to encode for a coding sequence (CDS; blue/orange). Self-contradictory CDS calls (orange); even automated annotation needs manual curation.

144. Comparing predictions in Artemis Glimmer(green)/Prodigal(blue) CDS prediction methods do not always agree (presence/absence, start position). How do we know which (if either) is best?

146. Using a “Gold Standard”: validationa a Pritchard and Broadhurst (2014) Methods Mol. Biol. 1127:53-64 doi:10.1007/978-1-62703-986-4 4 A general approach for all predictive methods • Deﬁne a known, “correct” set of true/false, positive/negative etc. examples - the “gold standard” • Evaluate your predictive method against that set for • sensitivity, speciﬁcity, accuracy, precision, etc. This ought to be done by the method developers, but often wise to evaluate in your own system. Many methods available, coverage beyond the scope of this introduction

147. Contingency Tables Condition (Gold standard) True False Test outcome Positive True Positive False Positive Negative False Negative True Negative Performance Metrics Sensitivity = TPR = TP/(TP + FN) Speciﬁcity = TNR = TN/(FP + TN) FPR = 1 − Speciﬁcity = FP/(FP + TN) If you don’t have this information, you can’t interpret predictive results properly.

148. “Gold Standard” results • Tested glimmer47 and prodigal48 on two enterobacterial close relatives as “gold standards” (still not perfect. . .) 1. Manually annotated (>3 expert person years) 2. Community-annotated (many research groups, interested in their own subset of genes) • Both methods trained directly on the annotated genes in each organism! 47 Delcher et al. (2007) Bioinformatics 23:673-679 doi:10.1093/bioinformatics/btm009 48 Hyatt et al. (2010) BMC Bioinf. 11:119 doi:10.1186/1471-2105-11-119

149. “Gold Standard” results Manually annotated: 4550 CDS genecaller glimmer prodigal predicted 4752 4287 missed 284 (6%) 407 (9%) Exact Prediction sensitivity 62% 71% FDR 41% 25% PPV 59% 75% Correct ORF sensitivity 94% 91% FDR 10% 3% PPV 90% 97%

150. “Gold Standard” results Community annotated: 4475 CDS genecaller glimmer prodigal predicted 4679 4467 missed 112 (3%) 156 (3%) Exact Prediction sensitivity 62% 86% FDR 31% 14% PPV 69% 86% Correct ORF sensitivity 97% 97% FDR 7% 3% PPV 93% 97%

151. Gene/CDS Prediction • Alternative CDS (and all other) prediction methods are unlikely to give identical results, or perform equally well • There is No Free Lunch (this is a theorem: http://en.wikipedia.org/wiki/No free lunch theorem) • To assess/choose between methods, performance metrics are required • Even on prokaryotes (a relatively simple case), current best methods for CDS prediction are imperfect • Manual correction is often required (usually the most demanding and time-consuming part of the process).

153. Prokaryotic Annotation Pipelinesa a Richardson and Watson (2012) Brief. Bioinf. 14:1-12 doi:10.1093/bib/bbs007 Many choices, including RAST49, PROKKA50, BaSYS51, etc. Often perform both CDS/feature calling and functional prediction. Two broad approaches: 1. Heavyweight: maintain database and resource, often annotating by homology, e.g. RAST 2. Lightweight: chain together multiple third-party packages, e.g. PROKKA Pipelines take a lot of tedium (and control) out of annotating bacterial genomes, but have the same issues as every other prediction tool. 49 Aziz et al. (2008) BMC Genomics 9:75 doi:10.1186/1471-2164-9-75 50 Seemann (2014) Bioinformatics 30:2068-2069 doi:10.1093/bioinformatics/btu153 51 Van Domselaar et al. (2005) Nuc. Acids Res. 33:W455-W459 doi:10.1093/nar/gki593

154. PROKKAa a Seemann (2014) Bioinformatics 30:2068-2069 doi:10.1093/bioinformatics/btu153 • Lightweight, and fast. • Runs locally. (5Mbp genome takes ≈10min on my desktop; more detailed ncRNA prediction takes ≈20min) • Flexible: built-in databases can be replaced by user databases. • Uses freely-accessible third-party tools for prediction Simple to run (at the command-line, or in Galaxy52). 52 Goecks et al. (2010) Genome Biol. 11:R86 doi:10.1186/gb-2010-11-8-r86

155. RASTa a Aziz et al. (2008) BMC Genomics 9:75 doi:10.1186/1471-2164-9-75 • Server-based (http://rast.nmpdr.org/). Queues likely. • Relies on SEED and FIGFam databases, held at NMPDR • FIGFam: isofunctional homologue families • Produces metabolic reconstruction

157. Principles of function prediction At genome scale, we realistically have to automate function prediction. Function prediction is just like any other prediction method. “Does this sequence imply that function?” Two main approaches to function prediction: • ab initio prediction (on basis of feature sequence/context only) • Unsupervised methods - not trained on an exemplar dataset • Supervised methods - trained on an exemplar dataset • homology matches (sequence similarity) • alignment to features with known/predicted functions

158. Homology-based function prediction Two proteins with similar sequence may have similar function. But. . . • How similar do they have to be (and where) to share the same function? • What do we mean by ‘same function’, anyway? Interaction/substrate speciﬁcity? Participation in a pathway? Contribution to a structure? Biochemical interconversion? . . . • How conﬁdent can we be in the comparator (annotated) sequence: was that function determined experimentally?

159. Gene Ontology (GO)a a Ashburner et al. (2000) Nat. Genet. 25:25-29 doi:10.1038/75556 The Gene Ontology provides a common vocabulary for describing biological function, and unifying functional descriptions. Ontologies (controlled vocabularies) are central to information-sharing. Gene Ontology Consortium: http://geneontology.org/ Many annotation tools and databases produce GO output, or compatible controlled vocabulary terms, e.g. • Blast2GO53: BLAST-based annotation • PHI-Base54: microbial pathogen-host interaction speciﬁc functions • GOPred55: combines several protein function classiﬁers 53 Conesa et al. (2005) Bioinformatics 21:3674-3676 doi:10.1093/bioinformatics/bti610 54 Winnenburg et al. (2006) Nuc. Acids Res. 34:D459-D464 doi:10.1093/nar/gkj047 55 Sarac et al. (2010) PLoS One 5:e12382 doi:10.1371/journal.pone.0012382

160. Gene Ontology (GO)a a Ashburner et al. (2000) Nat. Genet. 25:25-29 doi:10.1038/75556

161. Gene Ontology (GO)a a Ashburner et al. (2000) Nat. Genet. 25:25-29 doi:10.1038/75556

162. Are database annotations reliable?a a Schnoes et al. (2013) PLoS Comp. Biol. 9:e1003063 doi:10.1371/journal.pcbi.1003063 Are protein function annotations in databases determined experimentally, or by annotation transfer? High throughput experiments and genome annotations are conducted without validation of function, and placed in databases. • GO databases record annotation origin by publication • GO databases record evidence codes, e.g.: EXP=Inferred from Experiment; ISS=Inferred from Sequence Similarity • 0.14% of contributing publications provide 25% of all experimentally validated annotations in the Uniprot-GOA compilation. • There are biases in functional annotation. No clear solution to this kind of bias - but we have to recognise and account for it.

163. Are database annotations reliable?a a Radivojac et al. (2013) Nat. Meth. 10:221-227 doi:10.1038/nmeth.2340 The Critical Assessment of Function Annotation (CAFA) project.

164. Do biased database annotations matter? Experimental annotations of proteins are incomplete. But is that important? Tested by simulation, and following databases for three years.56 1. Yes. It matters. 2. Current large scale annotations are meaningful and almost surprisingly reliable. 3. The nature and level of data incompleteness, and type of classification model have an effect. 4. “Low precision, high recall” (i.e. less discriminating) tools most significantly affected. Molecular function prediction is usually more reliable than biological process prediction57 56 Jiang et al. (2014) Bioinformatics 30:i609-i616 doi:10.1093/bioinformatics/btu472 57 Cozzetto et al. (2013) BMC Bioinf. 14:S3-S1 doi:10.1186/1471-2105-14-S3-S1

165. CAFA resultsa a Radivojac et al. (2013) Nat. Meth. 10:221-227 doi:10.1038/nmeth.2340 The Critical Assessment of Function Annotation (CAFA) 2013 results. (F-measure combines precision and recall) • You can do better than BLAST. • Best-performing methods do comparably well. • Best methods used evolutionary relationships, structure, and expression data. • Machine Learning works best.

167. A wee trip to the doctor • You go for a checkup, and are tested for disease X • The test has sensitivity = 0.95 (predicts disease where there is disease) • The test has FPR = 0.01 (predicts disease where there is no disease)

168. A wee trip to the doctor • You go for a checkup, and are tested for disease X • The test has sensitivity = 0.95 (predicts disease where there is disease) • The test has FPR = 0.01 (predicts disease where there is no disease) • Your test is positive • What is the probability that you have disease X? • 0.01, 0.05, 0.50, 0.95, 0.99? • (Audience Participation!)

169. A wee trip to the doctor • What is the probability that you have disease X? • Unless you know the baseline occurrence of disease X, you cannot determine this.

170. A wee trip to the doctor • What is the probability that you have disease X? • Unless you know the baseline occurrence of disease X, you cannot determine this. • Baseline occurrence: fX • fX = 0.01 =⇒ P(disease|+ve) = 0.490 ≈ 0.5 • fX = 0.8 =⇒ P(disease|+ve) = 0.997 ≈ 1.0

172. Why Performance Metrics Mattera a Pritchard and Broadhurst (2014) Methods Mol. Biol. 1127:53-64 doi:10.1007/978-1-62703-986-4 4 • Imagine a paper describing a predictor for protein functional class (e.g. Type III eﬀector) • The paper reports sensitivity = 0.95, FPR = 0.01 • You run the predictor on 4,500 proteins in a new genome • It predicts 50 members of the class. How many of them are likely to be true positives?

173. Why Performance Metrics Mattera a Pritchard and Broadhurst (2014) Methods Mol. Biol. 1127:53-64 doi:10.1007/978-1-62703-986-4 4 • Imagine a paper describing a predictor for protein functional class (e.g. Type III eﬀector) • The paper reports sensitivity = 0.95, FPR = 0.01 • You run the predictor on 4,500 proteins in a new genome • It predicts 50 members of the class. How many of them are likely to be true positives? • We need a baseline level of that class (fX ) in the genome to determine this. • We estimate ≈ 45 members in protein complement, so fX = 0.01 • fX = 0.01 =⇒ P(class|+ve) = 0.490 ≈ 0.5

174. Bayes’ Theorem • May seem counter-intuitive: 95% sensitivity, 99% speciﬁcity =⇒ 50% chance of any prediction being incorrect • Probability given by Bayes’ Theorem • P(X|+) = P(+|X)P(X) P(+|X)P(X)+P(+| ¯X)P( ¯X)

175. Let’s play a game. . . 2, 4, 6, . . .

176. Bayes’ Theorem • May seem counter-intuitive: 95% sensitivity, 99% speciﬁcity =⇒ 50% chance of any prediction being incorrect • Probability given by Bayes’ Theorem • P(X|+) = P(+|X)P(X) P(+|X)P(X)+P(+| ¯X)P( ¯X) • This step commonly overlooked in the literature • conﬁrmation bias • people want to see positive examples/tell a story • people want to think their predictor works

177. A cautionary talea a Arnold et al. (2009) PLoS Pathog. 5:e1000376 doi:10.1371/journal.ppat.1000376 • Paper describes EffectiveT3, a type III eﬀector prediction tool • Reported sensitivity ≈ 0.71, FPR ≈ 0.15 • Applied tool to 739 complete bacterial and archaeal genomes

178. A cautionary talea a Arnold et al. (2009) PLoS Pathog. 5:e1000376 doi:10.1371/journal.ppat.1000376 • Paper describes EffectiveT3, a type III effector prediction tool • Reported sensitivity ≈ 0.71, FPR ≈ 0.15 • Applied tool to 739 complete bacterial and archaeal genomes • Organisms with an identifiable T3SS: 2-7% of genome predicted to be secreted • Organisms without an identifiable T3SS (or known not to have one): 1-10% of genome predicted to be secreted • “The surprisingly high number of (false) positives in genomes without T3SS exceeds the expected false positive rate” • This is not a surprise, statistically.

179. A cautionary talea a Arnold et al. (2009) PLoS Pathog. 5:e1000376 doi:10.1371/journal.ppat.1000376 Probability that an EffectiveT3 positive prediction corresponds to a secreted protein is given by Bayes’ Theorem • P(X|+) = P(+|X)P(X) P(+|X)P(X)+P(+| ¯X)P( ¯X) • P(+|X) = sensitivity = 0.71 • P(+| ¯X) = FPR = 0.15 • P(X) = base rate ≈ 0.03 (58) • =⇒ P(X|+) ≈ 0.13 Only 13% of predictions likely to be positive! How many predicted type III secreted proteins were there. . . 58 Boch and Bonas (2010) Annu. Rev. Phytopathol. 48:419-436 doi:10.1146/annurev-phyto-080508-081936

180. A cautionary talea a Arnold et al. (2009) PLoS Pathog. 5:e1000376 doi:10.1371/journal.ppat.1000376

181. Interpreting genome-scale predictionsa a Pritchard and Broadhurst (2014) Methods Mol. Biol. 1127:53-64 doi:10.1007/978-1-62703-986-4 4 • Statistics at genome-scale can be counterintuitive. • Use Bayes’ Theorem! • Predictions identify groups, not individual members of the group. e.g. • Test for airport smugglers has P(smuggler|+) = 0.9 • Test gives 100 positives • Which speciﬁc individuals are truly smugglers?

182. Interpreting genome-scale predictionsa a Pritchard and Broadhurst (2014) Methods Mol. Biol. 1127:53-64 doi:10.1007/978-1-62703-986-4 4 • Statistics at genome-scale can be counterintuitive. • Use Bayes’ Theorem! • Predictions identify groups, not individual members of the group. e.g. • Test for airport smugglers has P(smuggler|+) = 0.9 • Test gives 100 positives • Which speciﬁc individuals are truly smugglers? • The test does not allow you to determine this - you need more evidence for each individual • Same principle applies to other classiﬁers, (including protein functional class prediction) - watch for ‘cherry-picking’ in publications

184. Reconstructing metabolisma a Thiele and Palsson (2010) Nat. Protoc. 5:93-121 doi:10.1038/nprot.2009.203 Once metabolic functional annotation has been assigned to features, we can do comparative analysis of metabolism.

185. Dynamic models of metabolisma a Orth et al. (2010) Nat. Biotech. 28:245-248 doi:10.1038/nbt.1614 By using constraint-based models (e.g. Flux Balance Analysis), we can make these into dynamic representations of bacterial metabolism. • Upper, lower bounds to reaction rates • Deﬁne objective phenotype • Calculate conditions resulting in ﬂux • in silico knockouts

186. E. coli metabolisma a Monk et al. (2013) Proc. Natl. Acad. Sci. USA 110:20338-20343 doi:10.1073/pnas.1307797110 E. coli has a very long history of metabolic reconstruction59 Recent modelling work predicts which nutrients support growth 59 Reed and Palsson (2000) J. Bact. 185:2692-2699 doi:10.1128/JB.185.9.2692-2699.2003

187. E. coli metabolisma a Baumler et al. (2011) BMC Syst. Biol. 5:182 doi:10.1186/1752-0509-5-182 Models are complex, and experimental validation is essential There’s more we don’t know. . .

188. Microbial Genomics and Bioinformatics BM405 5.Finding Equivalent Features Leighton Pritchard1,2,3 1 Information and Computational Sciences, 2 Centre for Human and Animal Pathogens in the Environment, 3 Dundee Eﬀector Consortium, The James Hutton Institute, Invergowrie, Dundee, Scotland, DD2 5DA

191. What makes genome features equivalent? When we compare two features (e.g. genes) between two or more genomes, there must be some basis for making the comparison That is, they have to be equivalent in some way, such as: • common evolutionary origin • functional similarity • a family-based relationship It’s common to deﬁne equivalence of genome features in terms of evolutionary relationship.

192. Why look at equivalent features? The real power of genomics is comparative genomics! • Makes catalogues of genome components comparable between organisms • Diﬀerences, e.g. presence/absence of equivalents may support hypotheses for functional or phenotypic diﬀerence • Can identify characteristic signals for diagnosis/epidemiology • Can build parts lists and wiring diagrams for systems and synthetic biology

193. Evolutionary relationshipsa a Fitch (1970) Syst. Zool. 19:99-113 doi:10.2307/2412448 Equivalencies and relationships can be quite complex. We need precise terms to describe relationships between genome features. • analogy: functional similarity • homology: evolutionary common ancestor

195. Who let the -logues out?a a Fitch (2000) Trends Genet. 16:227-231 doi:10.1016/S0168-9525(00)02005-9 • homologues: elements that are similar because they share a common ancestor. There are NOT degrees of homology • analogues: elements that are (functionally?) similar, and this may be through common ancestry or some other means, e.g. convergent evolution • orthologues: homologues that diverged through speciation • paralogues: homologues that diverged through duplication within the same genome

196. Who let the -logues out?

201. ITYFIALMCTTa a Kristensen et al. (2011) Brief. Bioinf. 12:379-391 doi:10.1093/bib/bbr030 But it’s a little more complicated than that. Biology is not well-behaved. • Gene loss • Homologues may diverge so widely that they can be hard to recognise • Reconstructed evolutionary trees may not be robust inferences of speciation (or relevant to it, in prokaryotes) • There is no record of history - we can only make inferences All classiﬁcations of orthology/paralogy are inferences!

202. ITYFIALMCTTa a Kristensen et al. (2011) Brief. Bioinf. 12:379-391 doi:10.1093/bib/bbr030 All classiﬁcations of orthology/paralogy are inferences!

203. Ensembl Comparaa a Vilella et al. (2009) Genome Res. 19:327-335 doi:10.1101/gr.073585.107 Some tools/databases, e.g. Ensembl Compara, use slightly diﬀerent deﬁnitions (almost everything’s an “orthologue”)

205. Why focus on orthologues? Formalise the idea of corresponding genes in diﬀerent organisms. Orthologues serve two purposes: • Evolutionary equivalence • Functional equivalence (“The Ortholog Conjecture”60) Applications in comparative genomics, functional genomics and phylogenetics.61 Over 30 databases attempt to describe orthologous relationships (http://questfororthologs.org/orthology databases62) 60 Chen and Zhang (2012) PLoS Comp. Biol. 8:e1002784 doi:10.1371/journal.pcbi.1002784 61 Dessimoz (2011) Brief. Bioinf. 12:375-376 doi:10.1093/bib/bbr057 62 Altenhoﬀ and Dessimoz (2009) PLoS Comp. Biol. 5:e1000262 doi:10.1371/journal.pcbi.1000262

206. Finding orthologues Multiple methods and databases63,64,65 • Pairwise genome • RBBH (aka BBH, RBH), RSD, InParanoid, RoundUp • Multi-genome • Graph-based: COG, eggNOG, OrthoDB, OrthoMCL, OMA, MultiParanoid • Tree-based: TreeFam, Ensembl Compara, PhylomeDB, LOFT 63 Kristensen et al. (2011) Brief. Bioinf. 12:379-391 doi:10.1093/bib/bbr030 64 Trachana et al. (2011) Bioessays 33:769-780 doi:10.1002/bies.201100062 65 Salichos and Rokas (2011) PLoS One 6:e18755 doi:10.1371/journal.pone.0018755.g006

208. Which prediction methods work best? Taking advantage of prokaryotic operon structure: if the outer pair of a syntenic triplet of genes are orthologous, the middle gene is also likely to be orthologous.66 Speciﬁcally testing reciprocal best hits (RBH). 66 Wolf and Koonin (2012) Genome Biol. Evil. 4:1286-1294 doi:10.1093/gbe/evs100

209. Which prediction methods work best? • Tested on 573 prokaryotic genomes • 88-99% of RBH found in syntenic triplets • Overwhelming majority of middle genes are RBH RBH reliably ﬁnds orthologues.67 67 Wolf and Koonin (2012) Genome Biol. Evil. 4:1286-1294 doi:10.1093/gbe/evs100

210. Which prediction methods work best? Four methods tested against 2,723 curated orthologues from six Saccharomycetes • RBBH (and cRBH); RSD (and cRSD); MultiParanoid; OrthoMCL • Rated by statistical performance metrics: sensitivity, speciﬁcity, accuracy, FDR cRBH most accurate and speciﬁc, with lowest FDR.68 68 Salichos and Rokas (2011) PLoS One 6:e18755 doi:10.1371/journal.pone.0018755.g006

211. Which prediction methods work best? Testing on literature-based benchmarks for grouping by function and correct branching of phylogeny.69 69 Altenhoﬀ and Dessimoz (2009) PLoS Comp. Biol. 5:e1000262 doi:10.1371/journal.pcbi.1000262

212. Which prediction methods work best? • Performance varies by choice of method, and interpretation of “orthology” • Biggest inﬂuence is genome annotation quality • Relative performance varies with choice of benchmark • (clustering) RBH outperforms more complex algorithms under many circumstances

213. What is this magic RBH method?

215. Functional adaptation in Pbaa a Toth et al. (2006) Ann. Rev. Phytopath. 44:305-336 doi:10.1146/annurev.phyto.44.070505.143444

216. Functional adaptation in Pbaa a Toth et al. (2006) Ann. Rev. Phytopath. 44:305-336 doi:10.1146/annurev.phyto.44.070505.143444

218. Core genome Once equivalent genes have been identiﬁed, those present in all related isolates can be identiﬁed: the core genome. The core genome is expected to underpin common function. A core RBH cluster (clique) for 29 genomes:

219. Accessory genome The remaining genes are the accessory genome, and are expected to mediate function that distinguishes between isolates. An accessory RBH cluster for 29 genomes:

220. Accessory clusters Accessory RBH clusters can be pruned, to identify the accessory genome speciﬁc to subgroups of isolates: These genes may be responsible for subgroup-speciﬁc phenotypes

221. Accessory genome Accessory genomes act as a cradle for adaptive evolution70 This is particularly so for pathogens, such as Pseudomonas spp.71 70 Croll and Mcdonald (2012) PLoS Path. 8:e1002608 doi:10.1371/journal.ppat.1002608 71 Baltrus et al. (2011) PLoS Path. 7:e1002132 doi:10.1371/journal.ppat.1002132.t002

222. Core genome synteny Using tools like i-ADHoRe72 that identify synteny and collinearity, the structural organisation of the core genome can be determined: For Dickeya, the core genome appears to be structurally well-conserved across all isolates. 72 Proost et al. (2012) Nuc. Acids Res. 40:e11 doi:10.1093/nar/gkr955

223. Panseqa a Laing et al. (2010) BMC Bioinf. 11:461 doi:10.1186/1471-2105-11-461 Panseq is an online tool for identiﬁcation of core and accessory genomes, available at https://lfz.corefacility.ca/panseq/, and https://github.com/chadlaing/Panseq for standalone use

224. Harvesta a Treangen et al. (2014) Genome Biol. 15:524 doi:10.1186/s13059-014-0524-x Visualising and organising comparison/pangenome data across thousands of bacteria is diﬃcult. The Harvest suite of tools enables alignment and visualisation of thousands of genomes:

226. Things I didn’t get to

228. Conclusions

229. Conclusions

230. Conclusions

231. Licence: CC-BY-SA By: Leighton Pritchard This presentation is licensed under the Creative Commons Attribution ShareAlike license https://creativecommons.org/licenses/by-sa/4.0/

Microbial Genomics and Bioinformatics: BM405 (2015)

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Microbial Genomics and Bioinformatics: BM405 (2015) (20)

More from Leighton Pritchard (20)

Recently uploaded (20)

Microbial Genomics and Bioinformatics: BM405 (2015)