HiCanu: Resolving repeats and haplotypes

A new approach for assembly of HiFi accurate long reads

Transcript

1. Sergey Koren
   Staff Scientist, Genome Informatics Section, NHGRI
   Resolving repeats and haplotypes
   @sergekoren
   

   View Slide

2. What are CCS reads
   Number of reads
   99.9% 99.99%
   105,000
   90,000
   75,000
   60,000
   45,000
   30,000
   15,000
   Q20 Q25 Q30 Q35 Q40 Q45 Q50 Q55 Q60
   

   View Slide

3. Highly accurate overlaps, below 1%
   98.0
   98.5
   99.0
   99.5
   100.0
   Original RLE
   identity
   idy
   

   View Slide

4. <1% isn’t good enough
   5 10 20 50 100
   0 12 24 36 48 60 72 84
   Read length (kbp)
   NG50 (Mbp)
   >0.001% diverged
   >0.5% diverged
   Peregrine assemblies
   5 10 20 50 100
   0 12 24 36 48 60 72 84
   Read length (kbp)
   NG50 (Mbp)
   Chin et al. Human Genome Assembly in 100 Minutes. 2019
   

   View Slide

5. First trick: run-length encoding
   

   View Slide

6. Second trick: fix remaining errors
   

   View Slide

7. Last trick: ignore systematic errors
   

   View Slide

8. Perfect overlaps
   98.0
   98.5
   99.0
   99.5
   100.0
   Original RLE R
   identity
   idy
   Original RLE RLE + Correction
   identity
   name
   O
   R
   R
   

   View Slide

9. Near-optimal repeat resolution
   >0.001% diverged
   >0.5% diverged
   5 10 20 50 100
   0 12 24 36 48 60 72 84
   Read length (kbp)
   NG50 (Mbp)
   Peregrine assemblies
   HiCanu assemblies
   

   View Slide

10. Errors vs reference
    0
    100
    200
    300
    400
    500
    600
    CHM13 HG0733 NA12878
    Peregrine HiCanu Nanopore 10XG
    Structural errors vs GRCh38 by Quast
    Gurevich et al. QUAST: quality assessment tool for genome assemblies. 2013
    

    View Slide

11. Segmental duplication BAC resolution
    0%
    10%
    20%
    30%
    40%
    50%
    60%
    70%
    80%
    90%
    100%
    CHM13 HG0733 NA12878
    Peregrine HiCanu Nanopore 10XG
    30-fold 20 kbp 30-fold 100+ kbp
    Fraction of BAC bases correctly resolved
    

    View Slide

12. Remaining Collapsed Bases
    0.00
    50.00
    100.00
    150.00
    200.00
    250.00
    300.00
    CHM13 10kb CHM13 20kb
    Peregrine HiCanu Nanopore
    Q55
    Q48 Q27 Q58
    Q47 Q27
    Mbp with elevated coverage
    Vollger et al. Long-read sequence and assembly of segmental duplications. 2019
    

    View Slide

13. 1 2 3 4 5 6 7 8 9 10 11 12
    1 2 3 4 5 6 7 8 9 10 11 12
    HiFi and ONT UL are complementary
    30X CCS
    80X UL
    13 14 15 16 17 18 19 20 21 22 X 13 14 15 16 17 18 19 20 21 22 X
    

    View Slide

14. Haplotype separation improved too
    Canu HiFi
    Read N50: 10 kbp
    Total Block BP: 4.17 Gbp
    Phase block NG50: 362 kbp
    Switch 0.03%
    FALCON-Unzip CLR
    Read N50: 17 kbp
    Total block BP: 3.64 Gbp
    Phase block NG50: 229 kbp
    Switch: 0.15%
    Supernova 10XG
    Read N50: 95 kbp
    Total block BP: 4.64 Gbp
    Phase block NG50: 560 kbp
    Switch: 0.16%
    

    View Slide

15. • Applicable to any assembler
    • Approaching optimal repeat resolution
    • Your mileage will vary depending on repeat type
    • Centromeres still a challenge
    • Evidence of systematic error/coverage gaps
    • Human-level heterozygosity resolved
    • No polishing needed, consensus >Q50
    • Data collection is the bottleneck
    • 30 hr/cell, 5 days for a 3g mammal
    • CCS consensus, ≈6k cpu hrs, ≈ 2 days on 128 cores
    • Canu asm, ≈2k cpu hrs, ≈ 0.5 day on 128 cores (<8 hrs on a cluster)
    Conclusions
    

    View Slide

16. • Works on metagenomes: sheep rumen sample
    • 126 complete genomes (15% contaminated, 107 good)
    • Vs 61 for Canu 1.9 (5% contaminated, 58 good)
    • Vs 55 for Peregrine (11% contaminated, 49 good)
    • 41/44 circular marked by Canu complete and good
    • Vs 10/11 for Canu 1.9
    • More tweaks to come, lowering contamination rate
    One more thing
    Flye 2.5-g315122d: --meta --threads 56 --out-dir ccs_flye --genome-size 5m (1% error rate)
    Canu: genomeSize=3.1g 'batOptions=-eg 0.0 –sb 0.001 -dg 0 -db 3 -dr 0 -ca 2000 -cp 200’
    Canu 1.9: genomeSize=3.1g correctedErrorRate=0.015
    Peregrine: 24 24 24 24 24 24 24 24 24 --with-consensus --shimmer-r 3 --best_n_ovlp 8 --output asm
    

    View Slide

17. NHGRI
    • Sergey Nurk
    • Arang Rhie
    • Brian Walenz
    • Adam Phillippy
    • Evan E. Eichler
    • Mitchell Vollger
    • Glennis Logsdon
    • Robert Grothe
    • Jonas Korlach
    • Zev Kronenberg
    • Paul Peluso
    • David Rank
    • Kevin Fengler
    • Sung Bong Shin
    • Tim Smith
    Acknowledgements
    

    View Slide

18. View Slide