454 Nextgeneration sequencing Quality Control and Benchmarks

1
454 Next-generation sequencingQuality Control
and Benchmarks

• Jennifer M Taylor
• Bioinformatics Leader, Plant Industry
• June 2009

2
454 Quality Control

• Technical Details
• Roche 454 Workflow
• Applications
• Data
• General features
• Quality Metrics
• Genome Sequencing
• Variant Detection

3
Roche 454 Pyrosequencing

• High-throughput sequencing-by-synthesis
  technology
• GS FLX with Titanium series reagents
• Current 200-400 bp read lengths, 400 Mbp 10
  hour run
• Projected 500-700 bp read lengths (20kb, 8kb,
  3kb paired end reads)
• Applications
• Genome resequencing
• Variant Detection
• Denovo genome sequencing
• Metagenomics
• ChIPSeq protein-DNA interactions
• RNASeq transcript profiling small RNA
  profiling

4
Roche 454 Pyrosequencing

• Workflow
• Library preparation
• Fragmentation
• Adaptors
• Emulsion PCR
• Immobilisation
• Dilution to single bead / well
• Emulsion-based amplification

5
Roche 454 Pyrosequencing

• Workflow (contd)
• Sequencing by Synthesis
• Enrichment of bead with amplified fragments
• Arrayed to 1 bead / well / fibre-optic slide
• Nucleotides flows (T, A, C, G)
• Luciferase reaction measured for 1 or more
  nucleotide hybridisations
• Data Analysis
• Integration of position and signal
• Application of base-calling and quality filtering

6
Roche 454 General Data Features

• Flowgram / Pyrogram
• Does not read sequence base directly

Flow Cycles 3 Nucleotide Flows 12 Seq length
13
TCAGGTTTTTAAG
7
Roche 454 Pyrosequencing

• Sources of Error
• Undercall / Overcall
• T(0.49 , 0), T(1.6 , 2)
• Typically result in insertions / deletions
• Miscalls
• TCTTG True TCTCG (overcalled T AND
  undercalled C)
• Variation in sequence coverage
• PCR amplification bias
• PCR error (1)
• Sanger sequencing can average out PCR error
• NGS reads derived from a single molecule
  leading to transfer of error rate

8
Roche 454 Pyrosequencing
TCAGGTTTTTAAG TAACGGTTTACGG

• For sequence length (n)
• Min(f) n Max(f) 3n1
• n fixed f N(µ,s²)
• µ and s² increase linearly with sequence length
• Nucleotide frequencies ? equal µ ? max s² ? min
• f fixed n N(µ,s²)
• µ and s² increase linearly with flow cycle number
• Nucleotide frequencies ? equal µ ? min s² ? min

Kong, 2009
9
Roche 454 Quality Filters/Scores
0.5 0.55 0.6 0.65
0.7 0.75

• 0.5 lt signal lt 0.7 overlap region
• Allow only those reads lt5 of flows in overlap
  region
• Excluded reads trimmed from end until
• lt5 of flows in overlap region
• lt 82 flows (21 flow cycles)
• Exclude reads gt 5 ambiguous calls (N)

Margulies et al., 2005
10
Roche 454 Quality Filters/Scores

• n length of homopolymer
• s signal
• j is position in read
• P(sn) empirically determined to follow a
  Gaussian distribution
• P(n) for random nucleotide sequence is (¼)n

Margulies et al., 2005
11
454 Quality Control

• Technical Details
• 454 Workflow
• Applications
• Data
• General features
• Quality Metrics
• Genome Sequencing
• Variant Detection

12
Genome sequencing Accuracy and Coverage

• Alignment of multiple reads
• Integration with Sanger sequencing
• Wicker et al., 2006 Barley Genome
• Compared 454-derived and ABI-Sanger derived
  consensus sequences
• Error rates of 0.07 / position
• Moore et al., 2006 Plastid Genomes
• Comparison across genomes w.r.t consensus
• 0.031 0.043

13
Roche 454 Quality Filters/Scores

• Modifications to Quality Filters / Scores Huse
  et al., 2007
• 43 reference templates of known sequence
• Divergent bacteria
• gt 340,000 reads
• P(n) for random nucleotide sequence is (¼)n
• Penalises long homopolymer indiscriminately

14
Accuracy and Coverage

• Huse et al., 2007
• Error rate of 0.49
• 39 were homopolymer effects (insertions 36,
  deletions 27)
• lt 2 of reads accounted for nearly 50 of errors.
• Ambiguous bases
• Strong correlation between Ns and other types of
  errors
• 454 quality control allows up to 5 Ns per read
• Removal of all sequences containing Ns
• significant error rate improvement 0.24

15
Accuracy and Coverage

• Harismenday et al., 2009
• 260 kB across 4 individuals 454, Illumina, ABI
  Solid, ABI Sanger
• Saturating coverage 43 x (Roche), 188 x
  (Illumina), 841 x (SOLiD)

16
Accuracy and Coverage Variant Detection

• Harismenday et al., 2009
• Variant detection accuracy

ABI Sanger FP 0.9 FN 0.31
17
Accuracy and Coverage

• Amplicon End Bias
• 2.3 of total reference sequence
• 56 of Illumina sequence reads
• 11 of SOLiD
• 5 of 454
• Bias after fragmentation
• SOLiD and 454 library preparation adaptations
• Repeats
• SOLiD (1/2 fold coverage)
• 454 ( equal coverage)
• Illumina (2 fold coverage)
• Sequence composition
• Low coverage regions for SR tend to be AT rich.

18
Conclusions

• 454 Quality control scores need optimisation
• Naïve penalties for homopolymer length
• Inadequate control of ambiguities
• Lack of control of undercalling
• Brockman et al., 2008 Genome Research
• Uniformity of per-base sequence coverage needs to
  be improved
• Pyrosequencing shows high specificity in variant
  detection and low error rates in the construction
  of consensus sequencing IN THE PRESENCE OF
  saturating coverage.

19
Acknowledgements References

• Andrew Spriggs
• Karl Gordon
• David Townley
• David Lovell
• Brockman et al., Genome Research 2008, 18763-770
• Marguiles et al., Nature 2005, 437(15)376-380
• Kong, J. Comp. Biology 2009, 16(1)1-12
• Huse et al., Genome Biology 2007, 8R143
• Harismendy et al., Genome Biology 2009, 10R32

20
Thank you
Plant Industry Jennifer M Taylor Bionformatics
Leader Phone [phone] Email
Jen.Taylor_at_csiro.au Web