Taxonomic classification using Kraken 2

A tutorial on taxonomic classification with the Kraken suite, given at an ISMB 2026 tutorial session on July 12, 2026 at the Washington Hilton in Washington DC.

Transcript

1. Taxonomic classi fi cation using Kraken 2 Ben Langmead ISMB Tutorial, July 12, 9:45—10:45am 1 https://bit.ly/k-tut https://www.rcquinn.com/the-lincoln-overlooking-themall/ Washington Hilton, Washington DC, USA
2. Hi 👋 2 I’m a Professor of Computer Science at JHU I wrote some tools & resources you may have used like Bowtie, Bowtie 2, Movi, recount I DID NOT write Kraken, Derrick did! With Steven. I helped write Kraken 2; my group maintains Kraken, Kraken 2 Derrick Wood Steven Salzberg https://www.acentech.com/project/johns-hopkins-university-bloomberg-student-center/
3. Today ~35 mins of slide presentation ~12 mins of explorer ~8 mins of command line 3 https://bit.ly/k-tut All linked to on your tutorial page: Explorer Lecture Sandbox
4. Metagenomic sequencing Sequence all the DNA in an environment To characterize biodiversity To detect pathogens and map their spread To screen for new antibiotics and understand microbial resistance To study “microbiomes” on or in humans that a ff ect our health 4
5. Metagenomic sequencing Sequence all the DNA in an environment Dirt Hospital surface 💩 Stool Waterway etc 5
6. Metagenomic sequencing Sequence all the DNA in an environment Dirt Hospital surface Stool Waterway 6
7. Metagenomic sequencing Sequence all the DNA in an environment Dirt Hospital surface Stool Waterway 7
8. CCATAGTATATCTCGGCTCTAGGCCCTCATTTTTT GGGCTATATTAGGATCGCGCGTATGTACGGCTCGG ATATCTCGGCTCTAGGCCCTCATTTTTTTTAAATT GGGGTTATAGGATTTGCTTCGATTTTAGCTGCGTT Input DNA More details: Accurate whole human genome sequencing using reversible terminator chemistry. Nature. 2008 Nov 6;456(7218):53-9 Metagenomic sequencing Then… 8
9. Sequencing Synthesis (short reads) Most typical. High throughput, low cost, low error rates, mature tools. Similar competing techs are emerging, e.g. Avidity and SBX. Nanopore (long reads) Lower throughput & higher cost but MUCH longer reads. Good for distinguishing strains. Other PacBio HiFi, etc. Di ff erent cost/ accuracy/length trades. Also good for distinguishing strains 9
10. Whole-genome metagenomics sequencing “WGS”, “WMS”, etc WGS means that reads come from all over the genomes, not just one selected region or gene 📢. Not all metagenomics sequencing is WGS E.g. 16s RNA sequencing (doi: 10.1186/s40168-020-00900-2) 10
11. CGTCTGGGGGGTATGCACGCGATAGCATTGCGAGACGCTGGAGCCGGAGCACCCTATGTCGCAGTATCTGTCTTTGATTCCTG A genome GTATGCACGCGATAG TATGTCGCAGTATCT CACCCTATGTCGCAG GAGACGCTGGAGCCG Reads WGS 11
12. CGTCTGGGGGGTATGCACGCGATAGCATTGCGAGACGCTGGAGCCGGAGCACCCTATGTCGCAGTATCTGTCTTTGATTCCTG GTATGCACGCGATAG TATGTCGCAGTATCT CACCCTATGTCGCAG GAGACGCTGGAGCCG TAGCATTGCGAGACG GGTATGCACGCGATA TGGAGCCGGAGCACC CGCTGGAGCCGGAGC Reads A genome WGS 12
13. CGTCTGGGGGGTATGCACGCGATAGCATTGCGAGACGCTGGAGCCGGAGCACCCTATGTCGCAGTATCTGTCTTTGATTCCTG GTATGCACGCGATAG TATGTCGCAGTATCT CACCCTATGTCGCAG GAGACGCTGGAGCCG TAGCATTGCGAGACG GGTATGCACGCGATA TGGAGCCGGAGCACC CGCTGGAGCCGGAGC TGTCTTTGATTCCTG CGCGATAGCATTGCG GCATTGCGAGACGCT CCTATGTCGCAGTAT Reads A genome WGS 13
14. CGTCTGGGGGGTATGCACGCGATAGCATTGCGAGACGCTGGAGCCGGAGCACCCTATGTCGCAGTATCTGTCTTTGATTCCTG GTATGCACGCGATAG TATGTCGCAGTATCT CACCCTATGTCGCAG GAGACGCTGGAGCCG TAGCATTGCGAGACG GGTATGCACGCGATA TGGAGCCGGAGCACC CGCTGGAGCCGGAGC TGTCTTTGATTCCTG CGCGATAGCATTGCG GCATTGCGAGACGCT CCTATGTCGCAGTAT GACGCTGGAGCCGGA GCACCCTATGTCGCA GTATCTGTCTTTGAT CCTCATCCTATTATT TATCGCACCTACGTT CAATATTCGATCATG GATCACAGGTCTATC ACCCTATTAACCACT TGCATTTGGTATTTT CGTCTGGGGGGTATG CACGCGATAGCATTG GTATGCACGCGATAG ACCTACGTTCAATAT TATTTATCGCACCTA CCACTCACGGGAGCT GCGAGACGCTGGAGC CTATCACCCTATTAA CTGTCTTTGATTCCT ACTCACGGGAGCTCT CCTACGTTCAATATT GCACCTACGTTCAAT GTCTGGGGGGTATGC AGCCGGAGCACCCTA GACGCTGGAGCCGGA GCACCCTATGTCGCA GTATCTGTCTTTGAT CCTCATCCTATTATT TATCGCACCTACGTT CAATATTCGATCATG GATCACAGGTCTATC ACCCTATTAACCACT CACGGGAGCTCTCCA TGCATTTGGTATTTT CGTCTGGGGGGTATG CACGCGATAGCATTG CACGGGAGCTCTCCA Reads A genome WGS 14
15. Quantities matter More prevalence more reads → CCATAGTATATCTCGGCTCTAGGCCCTCATTTTTT… GGGCTATATTAGGATCGCGCGTATGTACGGCTCGG… ATATCTCGGCTCTAGGCCCTCATTTTTTTTAAATT… ATATCTCGGCTCTAGGCCCTCATTTTTTTTAAATT… ATATCTCGGCTCTAGGCCCTCATTTTTTTTAAATT… ATATCTCGGCTCTAGGCCCTCATTTTTTTTAAATT… GGGGTTATAGGATTTGCTTCGATTTTAGCTGCGTT… GGGGTTATAGGATTTGCTTCGATTTTAGCTGCGTT… TGCATTTGGTATTTT GTATGCACGCGATAG ACCTACGTTCAATAT GCGAGACGCTGGAGC CACGGGAGCTCTCCA TATCGCACCTACGTT CAATATTCGATCATG CACGGGAGCTCTCCA TGCATTTGGTATTTT TATCGCACCTACGTT CAATATTCGATCATG TGCATTTGGTATTTT GTATGCACGCGATAG ACCTACGTTCAATAT GCGAGACGCTGGAGC CTATCACCCTATTAA CCTACGTTCAATATT GCACCTACGTTCAAT GACGCTGGAGCCGGA GCACCCTATGTCGCA GGACGCACCTACGTT CAATATTCGATCATG CACGGGAGCTCTCCA TGCATTTGGTATTTT CACGGGAGCTCTCCA TATCGCACCTACGTT CAATATTCGATCATG TGCATTTGGTATTTT GTATGCACGCGATAG ACCTACGTTCAATAT GCGAGACGCTGGAGC CTATCACCCTATTAA CACGGGAGCTCTCCA 15
16. More on sequencing by synthesis 16 On your tutorial page https://bit.ly/k-tut
17. GTATGCACGCGATAG Input: reads from the environment TGCATTTGGTATTTT GTATGCACGCGATAG GCGAGACGCTGGAGC CACGGGAGCTCTCCA TATCGCACCTACGTT CAATATTCGATCATG CACGGGAGCTCTCCA TGCATTTGGTATTTT TATCGCACCTACGTT CAATATTCGATCATG TGCATTTGGTATTTT GTATGCACGCGATAG ACCTACGTTCAATAT GCGAGACGCTGGAGC CTATCACCCTATTAA CCTACGTTCAATATT GCACCTACGTTCAAT GACGCTGGAGCCGGA GCACCCTATGTCGCA GGACGCACCTACGTT CAATATTCGATCATG CACGGGAGCTCTCCA CACGGGAGCTCTCCA TATCGCACCTACGTT CAATATTCGATCATG TGCATTTGGTATTTT ACCTACGTTCAATAT GCGAGACGCTGGAGC CTATCACCCTATTAA CACGGGAGCTCTCCA TGCATTTGGTATTTT ACCTACGTTCAATAT 17
18. Output: each read classi fi ed to its point of origin TGCATTTGGTATTTT GTATGCACGCGATAG ACCTACGTTCAATAT GCGAGACGCTGGAGC CACGGGAGCTCTCCA TATCGCACCTACGTT CAATATTCGATCATG CACGGGAGCTCTCCA TGCATTTGGTATTTT TATCGCACCTACGTT CAATATTCGATCATG TGCATTTGGTATTTT GTATGCACGCGATAG ACCTACGTTCAATAT GCGAGACGCTGGAGC CTATCACCCTATTAA CCTACGTTCAATATT GCACCTACGTTCAAT GACGCTGGAGCCGGA GCACCCTATGTCGCA GGACGCACCTACGTT CAATATTCGATCATG CACGGGAGCTCTCCA TGCATTTGGTATTTT CACGGGAGCTCTCCA TATCGCACCTACGTT CAATATTCGATCATG TGCATTTGGTATTTT GTATGCACGCGATAG ACCTACGTTCAATAT GCGAGACGCTGGAGC CTATCACCCTATTAA CACGGGAGCTCTCCA 18
19. Taxonomy Possible “points of origin” are de fi ned by a taxonomy Taxon = named biological grouping, often a subtree of the tree of life Also: “Clade” Plural: taxa, clades 19 Image: Haeckel, Ernst. The Evolution of Man
20. Taxonomy Dengue virus type 1 Dengue virus type 2 Dengue virus type 3 Dengue virus Escherichia phage MS2 Encephalomyocarditis virus Enterovirus A Hepatovirus A Picornaviridae Orthornavirae Human immunodeficiency virus 1 Riboviria Escherichia phage Lambda Escherichia phage T7 Caudoviricetes Escherichia phage phiX174 Human papillomavirus 18 Monodnaviria Human mastadenovirus C Root We will work with this portion of the NCBI viral taxonomy today 20
21. Taxonomy Dengue virus type 1 Dengue virus type 2 Dengue virus type 3 Dengue virus Escherichia phage MS2 Encephalomyocarditis virus Enterovirus A Hepatovirus A Picornaviridae Orthornavirae Human immunodeficiency virus 1 Riboviria Escherichia phage Lambda Escherichia phage T7 Caudoviricetes Escherichia phage phiX174 Human papillomavirus 18 Monodnaviria Human mastadenovirus C Root For each node, you can look up its details in NCBI database We will work with this portion of the NCBI viral taxonomy today 21
22. Taxonomy Dengue virus type 1 Dengue virus type 2 Dengue virus type 3 Dengue virus Escherichia phage MS2 Encephalomyocarditis virus Enterovirus A Hepatovirus A Picornaviridae Orthornavirae Human immunodeficiency virus 1 Riboviria Escherichia phage Lambda Escherichia phage T7 Caudoviricetes Escherichia phage phiX174 Human papillomavirus 18 Monodnaviria Human mastadenovirus C Root We will work with this portion of the NCBI viral taxonomy today 22
23. Taxonomy Dengue virus type 1 Dengue virus type 2 Dengue virus type 3 Dengue virus Escherichia phage MS2 Encephalomyocarditis virus Enterovirus A Hepatovirus A Picornaviridae Orthornavirae Human immunodeficiency virus 1 Riboviria Escherichia phage Lambda Escherichia phage T7 Caudoviricetes Escherichia phage phiX174 Human papillomavirus 18 Monodnaviria Human mastadenovirus C Root In following slides, I use this even smaller excerpt 23
24. Taxonomy Dengue virus type 1 Dengue virus type 2 Dengue virus type 3 Encephalomyo carditis virus Enterovirus A Hepatovirus A Escherichia phage MS2 Picornaviridae Dengue virus Orthornavirae 24
25. Taxonomy Root D P E1 D1 D2 D3 P1 P2 P3 25
26. Refseq Root D P E1 D1 D2 D3 P1 P2 P3 26
27. The evidence: similarity to reference genomes >MT dna:chromosome chromosome:GRCh37:MT:1:16569:1 GATCACAGGTCTATCACCCTATTAACCACTCACGGGAGCTCTCCATGCATTTGGTATTTT CGTCTGGGGGGTATGCACGCGATAGCATTGCGAGACGCTGGAGCCGGAGCACCCTATGTC GCAGTATCTGTCTTTGATTCCTGCCTCATCCTATTATTTATCGCACCTACGTTCAATATT ACAGGCGAACATACTTACTAAAGTGTGTTAATTAATTAATGCTTGTAGGACATAATAATA ACAATTGAATGTCTGCACAGCCACTTTCCACACAGACATCATAACAAAAAATTTCCACCA AACCCCCCCTCCCCCGCTTCTGGCCACAGCACTTAAACACATCTCTGCCAAACCCCAAAA ACAAAGAACCCTAACACCAGCCTAACCAGATTTCAAATTTTATCTTTTGGCGGTATGCAC TTTTAACAGTCACCCCCCAACTAACACATTATTTTCCCCTCCCACTCCCATACTACTAAT CTCATCAATACAACCCCCGCCCATCCTACCCAGCACACACACACCGCTGCTAACCCCATA CCCCGAACCAACCAAACCCCAAAGACACCCCCCACAGTTTATGTAGCTTACCTCCTCAAA GCAATACACTGACCCGCTCAAACTCCTGGATTTTGGATCCACCCAGCGCCTTGGCCTAAA CTAGCCTTTCTATTAGCTCTTAGTAAGATTACACATGCAAGCATCCCCGTTCCAGTGAGT TCACCCTCTAAATCACCACGATCAAAAGGAACAAGCATCAAGCACGCAGCAATGCAGCTC AAAACGCTTAGCCTAGCCACACCCCCACGGGAAACAGCAGTGATTAACCTTTAGCAATAA ACGAAAGTTTAACTAAGCTATACTAACCCCAGGGTTGGTCAATTTCGTGCCAGCCACCGC GGTCACACGATTAACCCAAGTCAATAGAAGCCGGCGTAAAGAGTGTTTTAGATCACCCCC TCCCCAATAAAGCTAAAACTCACCTGAGTTGTAAAAAACTCCAGTTGACACAAAATAGAC TACGAAAGTGGCTTTAACATATCTGAACACACAATAGCTAAGACCCAAACTGGGATTAGA TACCCCACTATGCTTAGCCCTAAACCTCAACAGTTAAATCAACAAAACTGCTCGCCAGAA CACTACGAGCCACAGCTTAAAACTCAAAGGACCTGGCGGTGCTTCATATCCCTCTAGAGG AGCCTGTTCTGTAATCGATAAACCCCGATCAACCTCACCACCTCTTGCTCAGCCTATATA CCGCCATCTTCAGCAAACCCTGATGAAGGCTACAAAGTAAGCGCAAGTACCCACGTAAAG ACGTTAGGTCAAGGTGTAGCCCATGAGGTGGCAAGAAATGGGCTACATTTTCTACCCCAG AAAACTACGATAGCCCTTATGAAACTTAAGGGTCGAAGGTGGATTTAGCAGTAAACTAAG AGTAGAGTGCTTAGTTGAACAGGGCCCTGAAGCGCGTACACACCGCCCGTCACCCTCCTC AAGTATACTTCAAAGGACATTTAACTAAAACCCCTACGCATTTATATAGAGGAGACAAGT CGTAACCTCAAACTCCTGCCTTTGGTGATCCACCCGCCTTGGCCTACCTGCATAATGAAG AAGCACCCAACTTACACTTAGGAGATTTCAACTTAACTTGACCGCTCTGAGCTAAACCTA GCCCCAAACCCACTCCACCTTACTACCAGACAACCTTAGCCAAACCATTTACCCAAATAA AGTATAGGCGATAGAAATTGAAACCTGGCGCAATAGATATAGTACCGCAAGGGAAAGATG AAAAATTATAACCAAGCATAATATAGCAAGGACTAACCCCTATACCTTCTGCATAATGAA TTAACTAGAAATAACTTTGCAAGGAGAGCCAAAGCTAAGACCCCCGAAACCAGACGAGCT ACCTAAGAACAGCTAAAAGAGCACACCCGTCTATGTAGCAAAATAGTGGGAAGATTTATA GGTAGAGGCGACAAACCTACCGAGCCTGGTGATAGCTGGTTGTCCAAGATAGAATCTTAG TTCAACTTTAAATTTGCCCACAGAACCCTCTAAATCCCCTTGTAAATTTAACTGTTAGTC CAAAGAGGAACAGCTCTTTGGACACTAGGAAAAAACCTTGTAGAGAGAGTAAAAAATTTA ACACCCATAGTAGGCCTAAAAGCAGCCACCAATTAAGAAAGCGTTCAAGCTCAACACCCA CTACCTAAAAAATCCCAAACATATAACTGAACTCCTCACACCCAATTGGACCAATCTATC ACCCTATAGAAGAACTAATGTTAGTATAAGTAACATGAAAACATTCTCCTCCGCATAAGC CTGCGTCAGATTAAAACACTGAACTGACAATTAACAGCCCAATATCTACAATCAACCAAC AAGTCATTATTACCCTCACTGTCAACCCAACACAGGCATGCTCATAAGGAAAGGTTAAAA AAAGTAAAAGGAACTCGGCAAATCTTACCCCGCCTGTTTACCAAAAACATCACCTCTAGC ATCACCAGTATTAGAGGCACCGCCTGCCCAGTGACACATGTTTAACGGCCGCGGTACCCT CTCAAAGACCTGACCTTTGGTGATCCACCC-----GCCTNGGCCTTC |||||| |||| |||| ||||||||| |||| ||||| CTCAAACTCCTGGATTTTG--GATCCACCCAGCTGGCCTTGGCCTAA CTCAAACTCCTGACCTTTGGTGATCCACCCGCCTNGGCCTTC |||||||||||| ||||||||||||||||||||| ||||| | CTCAAACTCCTG-CCTTTGGTGATCCACCCGCCTTGGCCTAC Read Reference Read The higher the similarity, the better the chance it’s the true point of origin 27
28. How Kraken works We’ll start with Kraken 1 Wood, Derrick E., and Steven L. Salzberg. "Kraken: ultrafast metagenomic sequence classi fi cation using exact alignments." Genome Biology 15.3 (2014): R46. 28 k-mers
29. How Kraken works Shred read into k-mers: all overlapping length-k substrings We pick a particular value for k and stick with it throughout the analysis CCATAGTATATCTCGGCTCTAGGCCCTCATTTTTT CCAT GTAT TCTC GCTC AGGC CTCA TTTT CATA TATA CTCG CTCT GGCC TCAT TTTT ATAG ATAT TCGG TCTA GCCC CATT TAGT TATC CGGC CTAG CCCT ATTT AGTA ATCT GGCT TAGG CCTC TTTT Read: 4-mers: Typical settings for k are 21 — 31 (more on this later) 29
30. How Kraken works Kraken’s index takes k-mers and assigns them to taxa Images: Wood, Derrick E., and Steven L. Salzberg. "Kraken: ultrafast metagenomic sequence classi fi cation using exact alignments." Genome Biology 15.3 (2014): R46. Kraken index AKA “database” 30
31. How Kraken works Where do Kraken indexes come from? 🤔 or “databases” Build it yourself, with genomes from relevant taxa OR download a pre-built database Customizations also permitted; more on this later 31
32. If you know how a hash table works, you’re 90% there If you know how the index of a book works, you’re 75% there How Kraken’s index works… …not our focus for today but… 32
33. E1 D1 D2 D3 P1 P2 P3 Index is built over reference genomes for 🍁leaves🍁 of the taxonomic tree: How Kraken’s index works 🍁 🍁 🍁 🍁 🍁 🍁 🍁 33
34. E1 D1 D2 D3 P1 P2 P3 How Kraken’s index works 34 CCATAGTATATCTCGGCTCTAGGCCCTCATTTTTT CCAT GTAT TCTC GCTC AGGC CTCA TTTT CATA TATA CTCG CTCT GGCC TCAT TTTT ATAG ATAT TCGG TCTA GCCC CATT TAGT TATC CGGC CTAG CCCT ATTT AGTA ATCT GGCT TAGG CCTC TTTT Read: 4-mers: First we’ll talk about classifying k-mers, then we’ll talk about classifying reads
35. Root D P E1 D1 D2 D3 P1 P2 P3 Classifying k-mers 35 GTAT
36. D3 HIT Classifying k-mers GTAT 36
37. Root D P E1 D1 D2 D3 P1 P2 P3 Classifying k-mers 37 CCCT
38. Classifying k-mers 38 P2 HIT CCCT
39. P P1 P2 P3 HIT HIT HIT Classifying k-mers ATTT 39
40. Classifying k-mers We want to be speci fi c… …without being wrong 40
41. Root D P E1 D1 D2 D3 P1 P2 P3 Lowest common ancestor (LCA) LCA(node1, node2) = lowest node such that both node1 and node2 are at or below it 41
42. Lowest common ancestor (LCA) 42 LCA(D1, D2) = Root D P E1 D1 D2 D3 P1 P2 P3
43. Lowest common ancestor (LCA) 43 D D1 D2 LCA(D1, D2) = D
44. Lowest common ancestor (LCA) 44 LCA(P1, P2, P3) = Root D P E1 D1 D2 D3 P1 P2 P3
45. Lowest common ancestor (LCA) 45 P P1 P2 P3 LCA(P1, P2, P3) = P
46. Lowest common ancestor (LCA) 46 LCA(D2, E1) = Root D P E1 D1 D2 D3 P1 P2 P3
47. Lowest common ancestor (LCA) 47 LCA(D2, E1) = Root Root D E1 D2
48. Classifying k-mers For balance: assign k-mer to the LCA of the genomes in which it appears We want to be speci fi c… …without being wrong 48 ⚖
49. Classifying reads CCATAGTATATCTCGGCTCTAGGCCCTCATTTTTT CCAT GTAT TCTC GCTC AGGC CTCA TTTT CATA TATA CTCG CTCT GGCC TCAT TTTT ATAG ATAT TCGG TCTA GCCC CATT TAGT TATC CGGC CTAG CCCT ATTT AGTA ATCT GGCT TAGG CCTC TTTT Read: 4-mers: k-mers are the votes, the read is the election 49
50. D3 16 Classifying reads CCAT GTAT TCTC GCTC CATA TATA CTCG ATAG ATAT TCGG TAGT TATC CGGC AGTA ATCT GGCT 16 votes, all for D3 50
51. P2 16 Classifying reads CCAT GTAT TCTC GCTC CATA TATA CTCG ATAG ATAT TCGG TAGT TATC CGGC AGTA ATCT GGCT 16 votes, all for P2 51
52. Classifying reads 52 P1 P2 P3 6 5 5 CCAT TCTC GCTC CATA TATA CTCG ATAG ATAT TCGG TAGT TATC CGGC AGTA ATCT GGCT Votes split across P1, P2, P3 Now what? 🤔
53. Classifying reads Root D P E1 D1 D2 D3 P1 P2 P3 1 2 13 Somewhat more complicated split 🤔 🤔 53
54. Classifying reads Still more complicated split 🤔 🤔 🤔 54 Root D P E1 D1 D2 D3 P1 P2 P3 4 2 4 2 2 3 3
55. Classifying reads 55 LCA is the principle for assigning k-mers Once we’ve assigned k-mers, the principle for classifying the read is called the root-to-leaf path heuristic It has two parts: (a) sum up, (b) walk down Wood, Derrick E., and Steven L. Salzberg. "Kraken: ultrafast metagenomic sequence classi fi cation using exact alignments." Genome Biology 15.3 (2014): R46.
56. Classifying reads - sum up 56 Root D P E1 D1 D2 D3 P1 P2 P3 1 2 13 Sum up: for each node, sum all the k-mer assignments made at or below the node
57. Root D P E1 D1 D2 D3 P1 P2 P3 15 0 0 13 0 0 1 2 13 Classifying reads - sum up 57 Sum up: for each node, sum all the k-mer assignments made at or below the node Red #s are sums
58. Root D P E1 D1 D2 D3 P1 P2 P3 15 0 0 13 0 0 1 2 13 Classifying reads - walk down 58 Walk down: from root, keep moving to the child with largest sum. Stop on a tie.
59. Root D D1 D2 D3 15 13 0 0 1 2 13 Classifying reads - walk down Walk down: from root, keep moving to the child with largest sum. Stop on a tie. 59
60. Root D D1 15 13 1 2 13 Classifying reads - walk down Walk down: from root, keep moving to the child with largest sum. Stop on a tie. I win! 60
61. Classifying reads Root D P E1 D1 D2 D3 P1 P2 P3 1 2 9 4 61
62. Classifying reads 62 Root D P E1 D1 D2 D3 P1 P2 P3 15 0 0 9 4 0 1 2 9 4
63. Classifying reads 63 Root D D1 D2 D3 15 9 4 0 1 2 9 4
64. Classifying reads 64 Root D D1 15 9 1 2 9 I win! Even though there was a little evidence for D2!
65. Classifying reads 65 Root D P E1 D1 D2 D3 P1 P2 P3 8 0 8 4 2 0 3 0 3 20 4 2 4 2 2 3 3
66. Classifying reads 66 Root D P 8 8 4 2 2 I win!
67. Root D D1 D2 11 16 17 11 3 2 1 read has 20 k-mers — one box each no match clade(D1) = 11 → confidence = 11/20 = 0.55 clade(D) = 16 → confidence = 16/20 = 0.80 Con fi dence scores 67 Computes con fi dence score re fl ecting the proportion of votes supporting the answer
68. From Kraken to Kraken 2 Same principles, but Not all k-mers get a vote, only ones that are minimizers The hash table is probabilistic; makes mistakes with low, con fi gurable probability 68 Uses spaced seeds; slightly more robust to errors & di ff erences Wood, Derrick E., Jennifer Lu, and Ben Langmead. "Improved metagenomic analysis with Kraken 2." Genome biology 20.1 (2019): 257.
69. Time to start the explorer! 69 https://bit.ly/k-py
70. 70 Explorer: Input: Taxonomy
71. 71 Explorer: Input: Library
72. 72 Explorer: Input: Reads
73. 73 Explorer: Input: Classi fi cation
74. 74 Explorer: Input: Output record
75. To the sandbox! 75 https://bit.ly/k-sand Robert Aboukhalil
76. Recent activities 76 Partitioned indexes for easier building and querying, especially on tight memory budget Improved, modern wrapper script in Python (`k2`), with more robust connections to NCBI Integration of Genome Taxonomy Database (GTDB) and NCBI “core nt” databases
77. What bioinformaticians actually deal with If you analyze metagenomics data regularly, you will encounter issues with: Bizarre taxonomy decisions Contamination Low classi fi cation rate Need for non-default k-mer length 77 Papers on these topics are under “Further Reading” on your tutorial page: https://bit.ly/k-tut
78. Keep this link: 78 https://bit.ly/k-tut It can help on your ongoing journey
79. Mao-Jan Lin, Int. Ball. Cent. 7/13 12:40 (HiTSeq) Stephen Hwang, Je ff . West 7/15 17:20 (EvolCompGen) Nathaniel Brown, poster (HiTSeq) Please see my great students’ talks & posters Authors of Kraken, Kraken 2, KrakenUniq, Bracken 79 Derrick Wood Jen Lu Rone Charles Steven Salzberg Florian Breitwieser Peter Thielen Daniel Baker Support for Kraken 2 & Index zone: NIH R35GM139602 Amazon Web Services + me THANK YOU Ali & the organizers
80. Why not k-mers? 80 Picking k is not always easy; depends on: sequencing error rate density of sampled genomes assembly quality core versus accessory… If k-mer length becomes a puzzle in your work, keep an eye out for methods that use maximal exact matches or matching statistics instead The list goes on My group has some; see talks by Mao-Jan Lin, Stephen Hwang, posted by Nate Brown
81. Bracken If the goal is to quantify taxa, Kraken (1 or 2) can be combined with Bracken Bracken postprocesses read classi fi cations, estimates a quantity for each taxon 81 Lu, Jennifer, et al. "Bracken: estimating species abundance in metagenomics data." PeerJ Computer Science 3 (2017): e104. By looking at the “whole picture”, gives better estimates than Kraken 1/2 alone
82. KrakenUniq If the goal is to determine whether a particular taxon is present, use KrakenUniq Compiles information about how “well covered” each taxon is by the evidence (i.e. distinct k-mers matched) 82 Robust choice for pathogen detection / diagnosis of infection Breitwieser, Florian P., Daniel N. Baker, and Steven L. Salzberg. "KrakenUniq: con fi dent and fast metagenomics classi fi cation using unique k-mer counts." Genome Biology 19.1 (2018): 198. Kraken 2 has this built in!