Knowing Your NGS Upstream: Alignment and Variants

Knowing Your NGS
Upstream: Alignment
and Variants
March 27, 2013
Gabe Rudy, Vice President of
Product Development

Use the Questions pane in
your GoToWebinar window
Questions during
the presentation

Goals
 What I Assume About You
- Some experience with NGS technology
- Not a command line bioinformatician by day;
not afraid of technical terms
 What You Will Learn
- A healthy skepticism when looking at NGS
data
- What to expect/not expect from core labs or
upstream sequencing service providers
- Reading pile-ups in a genome browser and
spotting high quality vs sketchy variants
 What You Won’t Learn
- Interpreting biological significance of variants
- One true way to do secondary analysis
I won’t do this… hopefully!

My Background
 Golden Helix
- Founded in 1998
- Genetic association software
- Analytic services
- Hundreds of users worldwide
- Over 700 customer citations in scientific
journals
 Products I Build with My Team
- SNP & Variation Suite (SVS)
- SNP, CNV, NGS tertiary analysis
- Import and deal with all flavors of upstream data
- GenomeBrowse
- Visualization of everything with genomic coordinates.
All standardized file formats.
- RNA-Seq Pipeline
- Expression profiling bioinformatics

Agenda
Why You Should Care About Your Upstream?
Service Provider Deliverables: CEPH Trio Example
2
3
4
A Drink from the Bioinformatics Firehose
Background and Definitions1
Applications That Require Special Upstream Analysis5

NGS Analysis
Primary
Analysis
Secondary
Analysis
Tertiary
Analysis
“Sense Making”
 Analysis of hardware generated data, on-machine real-time stats.
 Production of sequence reads and quality scores
 QA and clipping/filtering reads
 Alignment/Assembly of reads
 Recalibrating, de-duplication, variant calling on aligned reads
 QA and filtering of variant calls
 Annotation and filtering of variants
 Multi-sample integration
 Visualization of variants in genomic context
 Experiment-specific inheritance/population analysis

Primary Analysis: I’d like some AGCT’s please
 Standardized on producing FASTQ
- AGCT or N
- Quality scores
- Pair of files for paired end
 Happens on machine for desktop
sequencers
- Ion Torrent processing microwell detectors
- MiSeq doing optic processing of flowcell
- PacBio processing optics of ZMW
 HiSeq 2000/2500
- Requires off-machine base-calling
- Can “call bases” with Illumina software on raw
data collected tile by tile

Assembly vs Alignment
 De Novo Genome Assembly
- Very difficult for large genomes to get
to “finished” genome quality
(traditionally done with Sanger).
- Short reads will get you to contigs
sizes of ~10-100Kb range.
- Need long reads (PacBio) or
restriction maps optical mapping
(OpGen) to make chromosomal sized
contigs
 Alignment
- Aligning to finished (or draft) genomes
that is considered “reference”
- Allows for some differences, but not
too many between your reads and the
reference

The Human Reference Sequence
 Genome Reference Consortium (GRCh37)
- Feb 2009, previous was NCBI36 March 2006
- 9 alt loci and 187 patches (11 patch releases)
 Supercontigs: Large unplaced contigs
- Some localized to chr level and some unknown
 Does not include a Mitochondrial reference
- UCSC hg19 includes older NCBI 36 MT
- 1000 genomes project using revised Cambridge
Reference Sequence (rCRS)
- Provide “g1k” reference: includes rCRS, Human
herpesvirus 4 type 1, supercontigs and “decoy”
sequence
 v38 genome coming this summer:
- Incorporate all patches into the reference
- Some allele fixes to have reference match major

Example Patch
 The tiling path in GRCh37 switched in the middle of ABO gene resulting in a
reference protein not present in humans.
 Patch adjusts tile path and fixes the problem.
 All patches will be incorporated into GRCh38, due this summer. Until then,
all alignment is done against unpatched reference.

Single Nucleotide Variants (i.e. SNVs or SNPs)
 Single base substitution from reference
 Note that “reference” is not always the “major” allele
 “Multi-allelic” sites have more than 2 cataloged
alleles
Gholson Lyon, 2012

Small Insertions/Deletions
 Generally defined as being < 150bp (often much shorter)
 Frameshift insertions/deletions important “loss of function” class of
variants
- Although InDels divisible by three are “in-frame” when in coding region
 Hard to call consistently. Poor concordance between algorithms.
 Where to call an InDel in a homopolymer?
- GTTTAC
- GTTTTAC
- 01234567
- How do you describe the insertion? Ins of T at 5? Or ins of T at 1?
- CGI in their v1 pipeline preferred calling insertion at end, others at beginning, now
always at beginning
 MNP – Can also be called differently

Copy Number Variants
 Required WGS
- CNVs > 10kb pretty accurate.
- 1kb to 10kb problematic.
 Detecting Deletions
- Can see coverage drop to near zero
- Harder to pinpoint breakpoint
- Possible false positives in low-
mappability regions
 Amplifications
- Can see coverage jump
- False positives due sample prep or
sequence artifacts
 Need “baseline,” look at Log Ratio
- Somatic detection uses normal
tissues
- Can have control population
Venter vs Watson WGS CNV-seq
64kbp gain in DDAH1 Gene of NA12878

Structural Variants
 Looking for:
- Balanced rearrangements
- Inversions
- Translocations
- Complex
 Signals to detect SV:
- Paired-end mappings too big (deletion)
- Depth of coverage
- Split-read mapping
 Translocations can result in “fusion”
genes.
- For example BCR-ABL fusion gene central in
pathogenesis certain leukemias.

Example 1kb Inversion (intron of APP)

Tertiary Analysis – “Sense Making”
 Detecting Known Clinically Relevant Variants
- Use targeted gene panels. Amplicons or custom capture.
- Look for carrier status or present of pathogenic or PGx variants
 Rare, Functional Variant Search and Interpretation
- Rare Mendelian Diseases
- Clinical Diagnostic: Ending Diagnostic Odyssey
- Looking for rare variants of functional consequence to a known phenotype
- Exome sequencing common, but whole genome has proponents
- Trios often used for looking at inheritance of putative variants (compound hets)
 Population Studies
- Like NHLBI or others studying complex disease
- Often looking at “variant burden” over genes between cases/controls
 Driver Somatic Variant Identification
- Looking for variants in tumor samples but not matched normal
- Not just SNPs and InDels, but CNVs and SVs

Not just DNA… but still DNA sequencing
 RNA-Seq
- Align to “transcriptome”, but often do analysis with reference genome coordinates and
reads “gapped” over introns they span in their spliced form
- Using read counts to approximate relative abundance of RNA in sample
- Compare relative abundance between groups
- Discover new transcribed genes or alternative splicing
 ChIP-Seq
- Measure sites and intensities of various proteins binding to DNA
- ENCODE project used to catalog TFBS and other functional elements
 Methyl-Seq
- Get sequences only with epigenetic methylation mark
- Run peak identification and intensity to look at relative levels of methylation

The Promise
 Both in research and clinical care, NGS is
powering discoveries making impactful
diagnoses
 Desktop sequencers and gene panels
much more economical than gene-by-gene
hunts
 Exomes have lead to many rare disease
diagnoses and affordably assay rare
functional variants
 Whole genomes have led to clinical
success stories and promise to be
instrumental to our understanding of complex
disease genetics
 Barrier to entry is lower than ever

Things That Can Confound Your Experiment
Library preparation errors Sequencing errors Analysis errors
 PCR amplification point
mutations (e.g. TruSeq
protocol, amplicons)
 Emultion PCR
amplification point
mutations (454, Ion
Torrent and SOLiD)
 Bridge amplification errors
(Illumina)
 Chimera generation
(particularly during
amplicon protocols)
 Sample contamination
 Amplification errors
associated with high or low
GC content
 PCR duplicates
 Base miscalls due to low
signal
 InDel errors (particular
PacBio)
 Short homopolymer
associated InDels (Ion
Torrent PGM)
 Post-homopolymeric tract
SNPs (Illumina) and/or
read-through problems
 Associated with inverted
repeats (Illumina)
 Specific motifs particularly
with older Illumina
chemistry
 Calling variants without
sufficient reads mapping
 Bad mapping (incorrectly
placed read)
 Correctly placed read but
InDels misaligned
 Multi-mapping to
repeat/paralogous regions
 Sequence contamination
e.g. adaptors
 Error in reference
sequence
 Alignment to ends of
contigs in draft assemblies
 Incorrect trimming of
reads, aligning adaptors
 Inclusion of PCR
duplicates
Nick Loman: Sequencing data: I want the truth! (You can’t handle the truth!)
Qual et al. A tale of three next generation sequencing platforms: comparison of Ion Torrent,
Pacific Biosciences and Illumina MiSeq sequencers. BMC Genomics. 2012 Jul

Your Choice of Technologies, Sometimes…
Platform Illumina MiSeq Ion Torrent PGM Ion Torrent Proton PacBio RS Illumina HiSeq 2000
Instrument Cost* $125 K $50 K $150K $695 K $654 K
Sequence yield per
run
1.5-2Gb
20-50 Mb on 314
chip, 100-200 Mb on
316, 1Gb on 318
10Gb on PI, 30GB on
PII (Mid 2013)
100 Mb 600Gb
Sequencing cost per
Gb*
$502 $500 (318 chip) $70 (PI chip) $2000 $41
Run Time 27 hours*** 2 hours 3 hours 2 hours 11 days
Reported Accuracy Mostly > Q30 Mostly Q20 Claimed >Q30 <Q10 Mostly > Q30
Observed Raw Error
Rate
0.80 % 1.71 % Probably ~1% 12.86 % 0.26 %
Read length up to 150 bases ~200 bp 100bp (200bp PII)
Average 1500
bases
up to 150 bases
Paired reads Yes Yes Yes No Yes
Insert size up to 700 bases up to 250 bases up to 250 bases up to 10 kb up to 700 bases
Typical DNA
requirements
50-1000 ng 100-1000 ng 100-1000 ng ~1 μg 50-1000 ng
Applications Targeted Targeted Exomes, RNA-Seq
Assembly,
Validation
Exomes, Genomes,
RNA
Qual et al. A tale of three next generation sequencing platforms: comparison of Ion Torrent,
Pacific Biosciences and Illumina MiSeq sequencers. BMC Genomics. 2012 Jul 29

[Show SNPs/Indels GenomeBrowse]

Agenda
2
3
4
Applications That Require Special Upstream Analysis5
 File formats
 Popular tools
 QA Filtering
 Visualization

FASTQ
 Contains 3 things per read:
- Sequence identifier (unique)
- Sequence bases [len N]
- Base quality scores [len N]
 Often “gzip” compressed (fq.gz)
 If not demultiplexed, first 4 or 6bp
is the “barcode” index. Used to
split lanes out by sample.
 Filtering may include:
- Removing adapters & primers
- Clip poor quality bases at ends
- Remove flagged low-quality reads
@HWI-ST845:4:1101:16436:2254#0/1
CAAACAGGCATGCGAGGTGCCTTTGGAAAGCCCCAGGGCACTGTGGCCAG
+
Y[SQORPMPYRSNP_][_babBBBBBBBBBBBBBBBBBBBBBBBBBB

SAM/BAM
 Spec defined by samtools author
Heng Li, aka Li H, aka lh3.
 SAM is text version (easy for any
program to output)
 BAM is binary/compressed version
with indexing support
 Alignment in terms of code of
matches, insertions, deletions,
gaps and clipping
 Can have any custom flags set by
analysis program (and many do)
Key Fields
 Chr, position
 Mapping quality
 CIGAR
 Name/position of mate
 Total template length
 Sequence
 Quality

VCF
 Specification defined by the 1000 genomes
group (now v4.1)
 Commonly compressed indexed with
bgzip/tabix (allows for reading directly by a
Genome Browser)
 Contains arbitrary data per “site” (INFO
fields) and per sample
 Single-Sample VCF:
- Contains only the variants for the sample.
 Multi-Sample VCF:
- Whenever one sample has a variant, all samples get
a “genotype” (often “ref”)
 Caveat:
- VCF requires a reference base be specified. Leaving
insertions to be “encoded” 1bp differently than they
are annotated
##fileformat=VCFv4.0
##fileDate=20090805
##source=myImputationProgramV3.1
##reference=1000GenomesPilot-NCBI36
##phasing=partial
##INFO=<ID=NS,Number=1,Type=Integer,Description="Number of Sa
##INFO=<ID=DP,Number=1,Type=Integer,Description="Total Depth"
##INFO=<ID=AF,Number=.,Type=Float,Description="Allele Frequen
##INFO=<ID=AA,Number=1,Type=String,Description="Ancestral All
##INFO=<ID=DB,Number=0,Type=Flag,Description="dbSNP membershi
##INFO=<ID=H2,Number=0,Type=Flag,Description="HapMap2 members
##FILTER=<ID=q10,Description="Quality below 10">
##FILTER=<ID=s50,Description="Less than 50% of samples have d
##FORMAT=<ID=GT,Number=1,Type=String,Description="Genotype">
##FORMAT=<ID=GQ,Number=1,Type=Integer,Description="Genotype Q
##FORMAT=<ID=DP,Number=1,Type=Integer,Description="Read Depth
##FORMAT=<ID=HQ,Number=2,Type=Integer,Description="Haplotype

Aligners
 BWA (also by Li H)
- Most prolifically used for genome alignment
- BWA-SW version geared for long reads (>100bp)
- Supports aligning with insertions/deletions to
reference
 Bowtie (John Hopkins, part of “Tuxedo” suite)
- Very fast, used commonly in RNA-Seq workflows
- Version 1 did not support “gapped” alignments
- Bowtie2 supports local gapped, longer reads
 Novoalign, Eland, SOAP, MAQ,
- Seed and expand strategy
 TopHat, SHRiMP, STAR, Gmap
- Specifically designed for ESTs
 Most improved by paired-end (mate-pairs)

InDel Re-Alignment
 Place, then realign “de Novo”
- Each read aligned independently by
global aligner.
- May have different preference of how to
handle “gaps” to reference.
 Local Re-Aligners for InDels
- Pindel
- GATK
 Important areas still problematic:
- CpG islands
- Promoter and 5′-UTR regions of the
genome
reference CAATC realignment CAATC
read1 CA-TC ----> CA-TC
read2 C-ATC CA-TC
Weixin Wang. Next generation sequencing has lower sequence coverage and poorer SNP-detection
capability in the regulatory regions. Scientific Reports 1, Article number: 55
AUC (area under the curve) comparison for different
genetic regions.

Variant Callers
 Samtools
- “mpileup” command computes BAQ, preforms local realignment
- Many filters can be applied to get high-quality variants
 GATK
- More than just a variant caller, but UnifiedGenotyper is widely used
- Also provides pre-calling tools like local InDel realignment and quality
score recalibration
 Custom tools specific to platform:
- CASAVA includes a variant caller for illumna whole-genome data
- Ion Torrent has a caller that handles InDels better for their tech

Quality Score Recalibration
DePristo (2011) A framework for variation discovery …. Nature Genetics 43(5) 491

BWA+GATK Best Practices Pipeline

Getting to High Confidence Variants
 Hard filters versus heuristic
based statistics
 <10bp considered threshold
for “low coverage”
 Quality score recalibration
Unfiltered Provided RD>10 & GQ>20 Exonic
SNPs 98621 89132 65009 19365
InDels 8141 7800 6503 428
Ts/Tv 2.36 2.45 2.54 3.26
Mendel Errors 234 202 46 3
Gabe
Trio

Ti/Tv
 Ts/Tv ratio can measure true
biological ratio of mutation types
versus sequence error:
- Random seq errors: 2/4 or 0.5
- Genome-wide: ~2.0-2.1
- Exome capture: ~2.5-2.8
- Coding: ~3.0-3.3
 Divergent too far than this
indicates random sequence
errors biasing the number.
DePristo (2011) A framework for variation discovery and genotyping using
next-generation DNA sequencing data. Nature Genetics 43(5) 491

Visualization
 Genome browsers:
- Validate variant calls
- Look at gene annotations,
problematic regions, population
catalogs
- Compare samples where no
variant called
 Free Genome Browsers:
- IGV
- Popular desktop by Broad
- UCSC
- Web-based, most extensive
annotations
- GenomeBrowse
- Designed to be publication ready
- Smooth zoom and navigation

Updates in Software Can Introduce Bugs
 Found 8K phantom variants in my
“final” 23AndMe exome

[My 23andMe “Buggy” Variant and Interpretation Example]

Recent Blog Post
http://blog.goldenhelix.com/?p=1725
The State of NGS Variant Calling: Don’t Panic!!

From which sequencing provider(s) do
you receive your upstream data from?
POLL:

Complete Genomics
 Different:
- Sequencing technology
- Alignment/variant calling algorithms
- File formats
 But high quality:
- MNPs, Indels
- CNV/SV calls
 Whole genome only
 Also provide tumor/normal pair
analysis
 Being acquired by BGI, some
question their sustainability

Complete Genomics Deliverables
 Summary statistics
 “var” and “masterVar” files
- Can be converted to VCF
- Some tools (like SVS) can import them
directly
 Evidence files
- Can be converted to BAM
 CNV, SV calls in text files
- CNV: Chr1:85980000-86006000 2.06 4x
gain, covers DDAH1
- SV: Chr21:27374158-27374699 common
inversion

Illumina Genome Network
 Standardized sequencing and
analysis, but multiple labs may be
contracted service provider.
 30x whole genomes
- SNPs, InDels, CNVs, SVs
- Concordance with SNP array (provided)
- Summary report
 Illumina provided tools used
- CASAVA toolkit with ELAND aligner
 Also provide Tumor/Normal pair
- Somatic SNVs and InDels identified by
looking at the tumor/normal together

Your Local Core Lab – Or 23andMe Exome Pilot!
 Research core labs often use a
BWA+GATK pipeline
- Especially for exomes
 Deliverables:
- VCF with SNVs, InDels
- BAM
 Tools for CNV/SV calling less
standardized
- Not commonly attempted with exomes

CEPH Trio
 The “benchmark” trio.
- Child, female NA12878 may be the most sequenced cell line
- Father NA12891 and Mother NA12892
 Whole Genome Data for Trio
- CGI with v2 pipeline
- IGN WGS at 30x, 100bp PE
- “Core Lab” BWA + GATK Best Practices on 100bp PE
 Concordance and Comparisons
- Lets interactively review examples where these three service providers differ and how.

GQ and DP of Shared de Novo Mutations

[deNovo and SV/CNV of NA12878 trio]

Agenda
2
3
4
Applications That Require Special Upstream
Analysis
5

Applications That Require Special Upstream Analysis
 MHC Region
 Somatic Variant Calling
 RNA-Seq
 Alu and other repeats
 Phased variants and complex MNP
 Moving to a new reference genome

Somatic Sniper and Friends
 Complete Genomics and IGN provide secondary alignment
specific to tumor/normal pairs.
 Do variant calling with on BAMs on pair in conjunction
 SomaticSniper approach:
 Covered by at least 3 reads
 Consensus quality of at least 20
 Called a SNP in the tumor sample with SNP quality of at least 20
 Maximum mapping quality of at least 40
 No high-quality predicted indel within 10 bp
 No more than 2 other SNVs called within 10 bp
 Not in dbSNP (non-cancer dbSNP)
 LOH filter (germline is het and tumor is homozygous)

What’s Next?
Download
www.goldenhelix.com
The Analysis and
Interpretation of My DTC
23andMe Exome
blog.goldenhelix.com
Killer App
Assembly and Alignment
Short Courses

Questions?
Use the Questions pane in
your GoToWebinar window

Knowing Your NGS Upstream: Alignment and Variants

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Knowing Your NGS Upstream: Alignment and Variants (20)

More from Golden Helix Inc (20)

Recently uploaded (20)

Knowing Your NGS Upstream: Alignment and Variants