Next-generation sequencing and SNPs was presented. The presentation discussed factors for finding real SNPs, including sequencing technology, mapping algorithms, variant calling, filtering, and experiences from exome resequencing. Accurate variant detection requires optimizing each step from sequencing to filtering false positives and negatives.

1. Next-generation sequencing
and SNPs
Jan Aerts
Wellcome Trust Sanger Institute
jan.aerts@gmail.com

2. Aim
To identify the SNP that causes disease,
phenotype
– Find them all, so you don’t miss it (false
negatives)
– Not find too many, so it’s useful (false
positives)

3. General principle
Map reads to reference sequence
Convert from read-based to base-based
(i.e. pileup)
Look at differences

4. This presentation
Factors in finding real SNPs
– Sequencing technology
– Mapping algorithms and initial calling
– Post-mapping tweaking
– Calling
– Filtering
Based on experiences in exome resequencing;
“experiment 5” on last slide Thomas

5. 1. Sequencing
• Provides raw data
• Different technologies
Different accuracy (critical!)
Different types of errors

6. Accuracy
Base quality drops
along read
Sanger
> SOLiD
> Illumina
> 454
> Helicos

7. Base calling errors
Main source of error for Illumina, less in
SOLiD & 454

8. Homopolymer runs
• Especially 454
39% of errors are homopolymers
• A5 motifs: 3.3% error rate
• A8 motifs: 50% error rate!
Reason: use signal intensity as a measure for
homopolymer length

10. Is it 4? Is it 5? Is it 4?

11. Consensus accuracy
Increase accuracy for SNP calling by
increasing coverage
– Illumina: 20X
– SOLiD: 12X
– 454: 7.4X
– Sanger: 3X
Factors: raw accuracy + read length

12. 2. Mapping: fastq => bam
• Maq and bwa: only 1 mapping
If multiple: mapQ = 0
<=> mosaik & mrFAST: alternatives
• Maq and bwa: use paired-end
information => might prefer correct
distance over correct alignment

13. 3. Post-mapping tweaking
Improve quality of mapped data:
– duplicate removal
– baseQ recalibration
– read clipping
– local realignment around indels
Genome Analysis Toolkit (GATK)
http://bit.ly/9zIn4b

14. Duplicate removal
PCR amplification bias
multiple reads with same start/stop =>
keep only one (with highest mapping Q)

15. java -Xmx2048m
-jar /path_to_picardtools/MarkDuplicates.jar
INPUT=input.bam
OUTPUT=output.bam
METRICS_FILE=output.metrics
VALIDATION_STRINGENCY=LENIENT
Picard
samtools
samtools rmdup input.bam output.bam

16. baseQ recalibration
• Why?
– correct for variation in quality with machine
cycle, sequence context, lane, baseQ…
• Steps:
– Identify what to correct for (create plots)
– Calculate covariates
– Apply covariates
– Check (create plots)

18. java -Xmx4g -jar GenomeAnalysisTK.jar
-l INFO
-R resources/Homo_sapiens_assembly18.fasta
--DBSNP resources/dbsnp_129_hg18.rod
-I my_reads.bam
-T CountCovariates
-cov ReadGroupCovariate
-cov QualityScoreCovariate
-cov DinucCovariate
-recalFile my_reads.recal_data.csv

19. java -Xmx4g -jar GenomeAnalysisTK.jar
-l INFO
-R resources/Homo_sapiens_assembly18.fasta
-I my_reads.bam
-T TableRecalibration
-outputBam my_reads.recal.bam
-recalFile my_reads.recal_data.csv

20. Read clipping
Remove:
– low quality strings of bases
– sections of reads
– reads containing user-provided sequences

21. Local realignment near indels

22. Local realignment near indels

23. java -Xmx1g -jar /path/to/GenomeAnalysisTK.jar
-T RealignerTargetCreator
-R /path/to/reference.fasta
-o /path/to/output.intervals
java -Xmx4g -Djava.io.tmpdir=/path/to/tmpdir
-jar /path/to/GenomeAnalysisTK.jar
-I input.bam
-R ref.fasta
-T IndelRealigner
-targetIntervals /path/to/output.intervals
-o realignedBam.bam

24. 4. SNP calling
• Different callers:
– samtools
– GATK UnifiedGenotyper
– SOAPsnp
–…
• Read-based => base-based

25. pileup
1 272 T 24 ,.$.....,,.,.,...,,,.,..^+. <<<+;<<<<<<<<<<<=<;<;7<&
1 273 T 23 ,.....,,.,.,...,,,.,..A <<<;<<<<<<<<<3<=<<<;<<+
1 274 T 23 ,.$....,,.,.,...,,,.,... 7<7;<;<<<<<<<<<=<;<;<<6
1 275 A 23 ,$....,,.,.,...,,,.,...^l. <+;9*<<<<<<<<<=<<:;<<<<
1 276 G 22 ...T,,.,.,...,,,.,.... 33;+<<7=7<<7<&<<1;<<6<
1 277 T 22 ..CCggC,C,.C.,,CC,..g. +7<;<<<<<<<&<=<<:;<<&<
1 278 G 23 ....,,.,.,...,,,.,....^k. %38*<<;<7<<7<=<<<;<<<<<
1 279 C 23 A..T,,.,.,...,,,.,..... ;75&<<<<<<<<<=<<<9<<:<<

26. pileup
1 272 T 24 ,.$.....,,.,.,...,,,.,..^+. <<<+;<<<<<<<<<<<=<;<;7<&
1 273 T 23 ,.....,,.,.,...,,,.,..A <<<;<<<<<<<<<3<=<<<;<<+
1 274 T 23 ,.$....,,.,.,...,,,.,... 7<7;<;<<<<<<<<<=<;<;<<6
1 275 A 23 ,$....,,.,.,...,,,.,...^l. <+;9*<<<<<<<<<=<<:;<<<<
1 276 G 22 ...T,,.,.,...,,,.,.... 33;+<<7=7<<7<&<<1;<<6<
1 277 T 22 ..CCggC,C,.C.,,CC,..g. +7<;<<<<<<<&<=<<:;<<&<
1 278 G 23 ....,,.,.,...,,,.,....^k. %38*<<;<7<<7<=<<<;<<<<<
1 279 C 23 A..T,,.,.,...,,,.,..... ;75&<<<<<<<<<=<<<9<<:<<

27. java
-Xmx6g
-jar /path_to/GenomeAnalysisTK.jar
-l INFO
-R human_b36_plus.fasta
-I input.bam
-T UnifiedGenotyper
--heterozygosity 0.001
-pl Solexa
-varout output.vcf
-vf VCF
-mbq 20
-mmq 10
-stand_call_conf 30.0
--DBSNP dbsnp_129_b36_plus.rod
GATK

28. samtools pileup
-vcs
-r 0.001
-l CCDS.txt
-f human_b36_plus.fasta
input.bam
output.pileup
samtools

29. VCF file
##fileformat=VCFv3.3
##FILTER=DP,"DP < 3 || DP > 1200"
##FILTER=QUAL,"QUAL < 25.0"
##FILTER=SnpCluster,"SNPs found in clusters"
##FORMAT=DP,1,Integer,"Read Depth"
##FORMAT=GQ,1,Integer,"Genotype Quality"
##FORMAT=GT,1,String,"Genotype"
##INFO=AB,1,Float,"Allele Balance for hets (ref/(ref+alt))"
##INFO=DB,0,Flag,"dbSNP Membership"
##INFO=DP,1,Integer,"Total Depth"
##INFO=HRun,1,Integer,"Largest Contiguous Homopolymer Run of Variant Allele In Either Direction"
##INFO=HaplotypeScore,1,Float,"Consistency of the site with two (and only two) segregating haplotypes"
##INFO=LowMQ,3,Integer,"3-tuple: <fraction of reads with MQ=0>,<fraction of reads with MQ<=10>,<total nubmer of reads>"
##INFO=MQ,1,Float,"RMS Mapping Quality"
##INFO=MQ0,1,Integer,"Total Mapping Quality Zero Reads"
##INFO=QD,1,Float,"Variant Confidence/Quality by Depth"
##annotatorReference=human_b36_plus.fasta
##reference=human_b36_plus.fasta
##source=VariantAnnotator
##source=VariantFiltration
#CHROM POS ID REF ALT QUAL FILTER INFO FORMAT a_a:bwa057_b:picard.bam
1 856182 rs9988021 G A 36.00 0;TARGET DB;DP=3;HRun=0;MQ=60.00;MQ0=0;QD=12.00;OnTarget=FALSE GT:DP:GQ 1/1:3:36.00
1 866362 rs4372192 A G 45.00 0;TARGET DB;DP=6;HRun=6;MQ=60.00;MQ0=0;QD=7.50;OnTarget=FALSE GT:DP:GQ 1/1:6:45.00
. . .

30. VCF file
##fileformat=VCFv3.3
##FILTER=DP,"DP < 3 || DP > 1200"
##FILTER=QUAL,"QUAL < 25.0" header
##FILTER=SnpCluster,"SNPs found in clusters"
##FORMAT=DP,1,Integer,"Read Depth"
##FORMAT=GQ,1,Integer,"Genotype Quality"
##FORMAT=GT,1,String,"Genotype"
##INFO=AB,1,Float,"Allele Balance for hets (ref/(ref+alt))"
##INFO=DB,0,Flag,"dbSNP Membership"
##INFO=DP,1,Integer,"Total Depth"
##INFO=HRun,1,Integer,"Largest Contiguous Homopolymer Run of Variant Allele In Either Direction"
##INFO=HaplotypeScore,1,Float,"Consistency of the site with two (and only two) segregating haplotypes"
##INFO=LowMQ,3,Integer,"3-tuple: <fraction of reads with MQ=0>,<fraction of reads with MQ<=10>,<total nubmer of reads>"
##INFO=MQ,1,Float,"RMS Mapping Quality"
##INFO=MQ0,1,Integer,"Total Mapping Quality Zero Reads"
##INFO=QD,1,Float,"Variant Confidence/Quality by Depth"
##annotatorReference=human_b36_plus.fasta
##reference=human_b36_plus.fasta
##source=VariantAnnotator
##source=VariantFiltration
#CHROM POS ID REF ALT QUAL FILTER INFO FORMAT a_a:bwa057_b:picard.bam
1 856182 rs9988021 G A 36.00 0;TARGET DB;DP=3;HRun=0;MQ=60.00;MQ0=0;QD=12.00;OnTarget=FALSE GT:DP:GQ 1/1:3:36.00
1 866362 rs4372192 A G 45.00 0;TARGET DB;DP=6;HRun=6;MQ=60.00;MQ0=0;QD=7.50;OnTarget=FALSE GT:DP:GQ 1/1:6:45.00
. . .
column header data

31. VCF file
INFO
DB;DP=3;HRun=0;MQ=60.00;MQ0=0;QD=12.00;OnTarget=FALSE
DB;DP=6;HRun=6;MQ=60.00;MQ0=0;QD=7.50;OnTarget=FALSE
FORMAT a_a:bwa057_b:picard.bam
GT:DP:GQ 1/1:3:36.00
GT:DP:GQ 1/1:6:45.00

32. Pileup => VCF
Custom scripts, then annotate
java -Xmx10g
-jar GenomeAnalysisTK.jar
-T VariantAnnotator
--assume_single_sample_reads sample
-R human_b36_plus.fasta
-D dbsnp_129_b36_plus.rod
-I input.bam
-B variant,VCF,unannotated.vcf
-o annotated.vcf
-A AlleleBalance
-A MappingQualityZero
-A LowMQ
-A RMSMappingQuality
-A HaplotypeScore
-A QualByDepth
-A DepthOfCoverage
-A HomopolymerRun

33. 5. Filtering
• Aim: to reduce number of false positives
• Options:
– Depth of coverage
– Mapping quality
– SNP clusters
– Allelic balance
– Number of reads with mq0

34. java
-Xmx4g
-jar GenomeAnalysisTK.jar
-T VariantFiltration
-R human_b36_plus.fasta
-o output.vcf
-B variant,VCF,input.vcf
--clusterWindowSize 10
--filterExpression 'DP < 3 || DP > 1200'
--filterName 'DP'
--filterExpression 'QUAL < #{qual_cutoff}'
--filterName 'QUAL'
--filterExpression 'AB > 0.75 && DP > 40'
--filterName 'AB'

35. Filtering - QC metrics (1)
Transition/transversion ratio
Random: Ti/Tv = 0.5
Whole genome: 2.0-2.1
Exome: 3-3.5

36. Filtering - QC metrics (2)
Number of novel SNPs
Exome:
total 20k - 25k;
novel 1-3k

38. Combining discovery pipelines
• Mapper: MAQ/bwa/stampy/…
• BaseQ recalibration? Local
realignment?
• SNP caller: GATK/samtools/SOAPsnp
• Priors for SNP calling: heterozygosity
(whole genome, exome, dbSNP)
• Filtering

39. Combining discovery pipelines
true positives
ROC
false positives

40. Combining discovery pipelines
combinations
single
better

41. Indels
Still more tricky than SNPs
– samtools/dindel/GATK
– Sample of 10 individuals: on average per
individual:
• 2 novel functional high-quality SNPs
• 18 novel functional high-quality indels
“I trust manual interpretation of the reads more
than the basic quality parameters we use”

42. 4 snp_1 STOP_GAINED
1 snp_2 STOP_LOST
1 snp_3 STOP_GAINED
1 snp_4 ESSENTIAL_SPLICE_SITE,INTRONIC
1 snp_5 ESSENTIAL_SPLICE_SITE,INTRONIC
2 snp_6 STOP_GAINED
2 snp_7 STOP_GAINED
1 snp_8 STOP_GAINED
1 snp_9 STOP_GAINED
1 snp_10 STOP_GAINED
1 snp_11 STOP_GAINED
1 snp_12 STOP_GAINED
1 snp_13 STOP_LOST
1 snp_14 STOP_GAINED

43. 4 snp_1 STOP_GAINED
1 snp_2 STOP_LOST
1 snp_3 STOP_GAINED
1 snp_4 ESSENTIAL_SPLICE_SITE,INTRONIC
1 snp_5 ESSENTIAL_SPLICE_SITE,INTRONIC
2 snp_6 STOP_GAINED
2 snp_7 STOP_GAINED
1 snp_8 STOP_GAINED
1 snp_9 STOP_GAINED
1 snp_10 STOP_GAINED
1 snp_11 STOP_GAINED
1 snp_12 STOP_GAINED
1 snp_13 STOP_LOST
1 snp_14 STOP_GAINED
178 indels FRAMESHIFT_CODING

44. Conclusions
Different tools exist and are created
Best to combine (intersect) the results from
different pipelines
Genome Analysis ToolKit (GATK) provides
useful bam-file processing tools:
– Realignment around indels
– Base quality recalibration

45. Use in resequencing
• Identify SNPs/indels
• Consequences (loss-of-function?)
• Prevalence in cases/controls
• Model:
– Dominant: any het
– Recessive: homnonref or compound het

46. References
• Chan E. In: Single Nucleotide Polymorphisms,
Methods in Molecular Biology 578 (2009)
• McKenna et al. Genome Res 20:1297-1303 (2010)
• Li H & Durbin R. Bioinformatics 25:1754-1760 (2009)
• Li H et al. Bioinformatics 25:2078-2079 (2009)
• Li H et al. Genome Res 18:1851-1858 (2008)

47. Questions?