talks

GenotypeRefinementWorkflow

Usingaddi8onaldatatoimprovegenotypecallsandlikelihoods

YouarehereintheGATKBestPrac8cesworkflowforgermlinevariantdiscovery

Analysis-Ready Variants

111Raw Reads

Raw Variants IndelsSNPs

Analysis-ReadyReads

Indel Realignment

Base Recalibration

SNPs & Indels

Variants

IndelsSNPs

VariantAnnotation

Variant Evaluation

look good?

use in projecttroubleshoot

111Analysis-ReadyReads

Genotype Likelihoods

Joint Genotyping

Analysis-Ready

No

n-G

AT

K

Mark Duplicates& Sort (Picard)

Var. Calling HC in ERC mode

separately per variant type

Variant Recalibration

Map to Reference

BWA mem GenotypeRefinement

Data Pre-processing Variant Discovery>> >> Callset Refinement

Whycareaboutgenotypes?

•  Medicalgene8cistsneedgenotypesforpa8ents–  Doanypa8entshavetwocopiesofaLOFmuta8on?–  Aretheparentsofadiseasedchildlikelytohavemoreafflictedchildren?

•  Popula8ongene8cistsneedgenotypesforassocia8onstudies–  Howdoesthenumberofcopiesofanalleleaffectthephenotype?

Variantcallvs.Genotypecall

•  Variantcall=thereisvaria8onatthissiteinoneormoresamples

•  Genotypecall=thesearethehaplotypespresentinthissample

Genotypecallqualityisimportant!

•  Somesites/sampleshavepoorgenotypecalls–  Canbeambiguousduetolowconfidence–  Mightbeen8relywrong!

•  Canaddi8onal(independent)dataimprovegenotypecalls?–  Usehighqualitydata(like1000G)aspriors–  Usepedigree(ifavailable)–  Calculateposteriorgenotypeprobabili8es

ReviewofBayes’sRule

Giventhatyourcoworkerjustwalkedinwithanumbrella,whatistheprobabilitythatitisraining?•  Observa8on=umbrella•  Θ=probabilityofrain

prior

(normalize)

likelihoodposteriorprobability

GenotypeRefinementWorkflow

High Quality Variants

De Novo Variants

Recalibrated Variants

Population Priors Family Priors

CalculateGenotypePosteriors

VariantFiltration

VariantAnnotator

Variants with Posterior Qualities

CalculateGenotypePosteriors

java -jar GenomeAnalysisTK.jar \ -T CalculateGenotypePosteriors \ -R reference.fasta \ -V input.vcf \ -ped family.ped \ -supporting population.vcf \ -o output.vcf

High Quality Variants

De Novo Variants

Recalibrated Variants

Population Priors Family Priors

CalculateGenotypePosteriors

VariantFiltration

VariantAnnotator

Variants with Posterior Qualities

Case1:HOM_VARCallw/LowFrequencyPriors

1) Baseline HOM_VAR call

2) Priors w/low allele frequency applied

3) Posterior genotype called HET

4) In agreement w/NISTand BAMs

Likelihoods x Priors = Posterior Probabilities[895,3,0] AF=0.002 [868,0,27]

[HOM_REF, HET, HOM_VAR] [HOM_REF, HET, HOM_VAR]

Genotype correctedConfidence improved from Q3 to Q27

Case2:HETCallwithHighFrequencyPriors

1) Baseline HET call

2) Priors w/high allele frequency applied

3) Posterior genotype called HOM_VAR

4) In agreement w/NISTand BAMs

Likelihoods x Priors = Posterior Probabilities[894,0,0] AF=0.987 [932,16,0]

[HOM_REF, HET, HOM_VAR] [HOM_REF, HET, HOM_VAR]

Genotype corrected Confidence improvedfrom Q0 to Q16

Popula8onpriorsimprovegenotypeconfidence

Baseline HomRef calls are under confident, but posterior calls are more accurate

Baseline HomVar calls are over confident, but posterior calls are improved

HomRef

HetHom

Var

Assessingconfidenceandcorrectness

Intercept = 9.9612Slope = 0.9302

Average Q10 increase for correct calls, Q≤30

BaselineGenotypeQuality

PosteriorG

enotypeQuality

Incorrect calls stay about the same

HomozygousReferenceCalls

Parentalgenotypesinformchildgenotypes

!(!!|!! ,!! ,!!)!

Child Mother FatherHR HR HRHR HR HETHR HET HRHR HET HET

Child Mother FatherHET HET HETHET HR HETHET HET HRHET HV HETHET HET HVHET HR HVHET HV HR

Child Mother FatherHV HV HVHV HV HETHV HET HVHV HET HET

•  Childcanonlyinheritallelespresentinparents•  Parentgenotypesdeterminepossiblechildgenotypes

(assumingnomuta8ons)

•  HaplotypeCallergives•  Giventriodatawecanderive

! ! !"# = ! ! !"# ! !(!)!(!"#|!)!(!)!

!

Bayesianpriorsappliedtotrios

•  RecallBayes’sRule:

•  Establishgenotypeconfigura8onprobabili8es

•  Applyfamilypriors

! !! = !" ! = !!(!! = !") !! !! !! !! !(!)!!,!!!(! ! ! !(!) !

applyprior

normalize

likelihoodposterior

! !!,!! ,!! = !! !!, 1!"!!, 2!"#

1− 10! − 2!!,!"! −!"!

!!

Familypriorsimprovegenotypeconfidence

Baseline HomRef calls are under confident, but posterior calls are more accurate

Posterior HomRef and HomVar calls are higher confidence

HomRef

HetHom

Var

Assessingconfidenceandcorrectness

Intercept = 12.831,Slope = 1.238

Average Q13 increase for correct calls, Q≤30

BaselineGenotypeQuality

PosteriorG

enotypeQuality

Incorrect calls stay about the same

HomozygousReferenceCalls

FilterlowconfidenceGQs

•  UseVariantFiltra8ontofilterambiguous,low-confidencecalls

•  RecommendedthresholdisGQ=20–  GQ20isPhred-scaled99%confidence

•  Restrictfurtheranalysistohigh-qualitydata

High Quality Variants

De Novo Variants

Recalibrated Variants

Population Priors Family Priors

CalculateGenotypePosteriors

VariantFiltration

VariantAnnotator

Variants with Posterior Qualities java -jar GenomeAnalysisTK.jar \ -T VariantFiltration \ -R reference.fasta \ -V input.vcf \ --filterExpression “GQ<20” \ --filterName “lowGQ” \ -o output.vcf

VariantAnnotator

High Quality Variants

De Novo Variants

Recalibrated Variants

Population Priors Family Priors

CalculateGenotypePosteriors

VariantFiltration

VariantAnnotator

Variants with Posterior Qualities

java -jar GenomeAnalysisTK.jar \ -T VariantAnnotator \ -R reference.fasta \ -V input.vcf \ -A PossibleDeNovo \ -o output.vcf

WhatareDeNovomuta8ons?

•  CulpritsinmanyrareMendeliandisorders•  ~30denovomuta8onsoccurperhumangenome

Parentsarehomozygousreference

Childishet(onecopyofaltallele)

Proper8esofsequencedDeNovos

•  Novelty–  Childhasonlyaltalleleintrio,notinherited

•  Rarity–  Allelefrequencyacrossallsamplessequencedislow

•  Confidence–  SetGQthresholdforparentsandchild–  (GQimprovementtoolshelpALOThere!)

Exampleofaclinicalcase

•  Realclinicaldata•  Suspecteddenovomuta8oninoffspring

417denovosfromrawGTcalls

17denovosbasedonposteriorGTs

8highconfidencedenovosanerGQfiltering

Priorscanbetunedforsensi8vity

Sensi8vityandspecificitycanbetunedasinVQSR

Muta8onpriorisaparameteringenotypeconfigura8onprobability:

IncreasingSensi8vity

! !!,!! ,!! = !! !!, 1!"!!, 2!"#

1− 10! − 2!!,!"! −!"!

!!

Genotyperefinementyieldsmorehigh-qualitygenotypes

•  Ini8algenotypecallsmaybeambiguousorwrong

•  Applyingpopula8on+familypriorsimprovesconfidence

•  Morehighconfidencegenotypes->moredatafordownstreamanalysis!Hom

Var

YouarehereintheGATKBestPrac8cesworkflowforgermlinevariantdiscovery

Analysis-Ready Variants

111Raw Reads

Raw Variants IndelsSNPs

Analysis-ReadyReads

Indel Realignment

Base Recalibration

SNPs & Indels

Variants

IndelsSNPs

VariantAnnotation

Variant Evaluation

look good?

use in projecttroubleshoot

111Analysis-ReadyReads

Genotype Likelihoods

Joint Genotyping

Analysis-Ready

No

n-G

AT

K

Mark Duplicates& Sort (Picard)

Var. Calling HC in ERC mode

separately per variant type

Variant Recalibration

Map to Reference

BWA mem GenotypeRefinement

Data Pre-processing Variant Discovery>> >> Callset Refinement

talks

Furtherreading

hrp://www.broadins8tute.org/gatk/guide/

hrps://www.broadins8tute.org/gatk/gatkdocs/org_broadins8tute_gatk_tools_walkers_variantu8ls_CalculateGenotypePosteriors

hrp://www.broadins8tute.org/gatk/guide/ar8cle?id=4723

hrp://www.broadins8tute.org/gatk/guide/ar8cle?id=4726

talks