Calling DNA Variants Steve Laurie Centro Nacional de Analisis Genomico (CNAG-CRG), Barcelona

Transcription

1 Calling DNA Variants Steve Laurie Centro Nacional de Analisis Genomico (CNAG-CRG), Barcelona Variant Effect Predictor Training Course Heraklion, 31st October 2016

2 Calling DNA Variants - Overview What is a variant, and how many do we have? What do we do with our lab coats on? What do we do sitting at a computer? Some real world issues How well do we do in general?

3 What is a variant? 3 A variant is any position/region in our sample which differs from the haploid reference genome to which we are comparing it Single Nucleotide Variants (SNVs) e.g. A G note diploid individual may be AA, AG, or GG Short insertions and deletions (InDels) e.g. TA TATA (insertion of TA ) e.g. CT C (loss of the T at the third position) Copy Number Variants (CNVs) tandem duplication of longer regions (~1-100kb) that are typically polymorphic within the population e.g. AMY1 Structural Variants (SVs) larger still, and often more complex

4 What is a variant? 4 A variant is any position/region in our sample which differs from the haploid reference genome to which we are comparing it Single Nucleotide Variants (SNVs) e.g. A G note diploid individual may be AA, AG, or GG Short insertions and deletions (InDels) e.g. TA TATA (insertion of TA ) e.g. CT C (loss of the T at the third position) Copy Number Variants (CNVs) tandem duplication of longer regions (~1-100kb) that are typically polymorphic within the population e.g. AMY1 Structural Variants (SVs) larger still, and often more complex

5 How many variants do we have? 5 A variant is any position/region in our sample which differs from the haploid reference genome to which we are comparing it Single Nucleotide Variants (SNVs) ~ 3,500,000 4,000,000 ( ~ 30, ,000 exomic ) Short insertions and deletions (InDels) ~ 300, ,000 Copy Number Variants (CNVs) ~ 5-10% of genome Structural Variants larger still, and often more complex ~ 13% of genome

6 NGS Workfl ow what do we do with our lab coats on? 6 Kassahn, K. (2013)

7 NGS Workfl ow what do we do with our lab coats on? 7 Kassahn, K. (2013)

8 Targeted NGS Fracti onati on and Capture fractionation Figure from Mardis, E.R. (2012)

9 Targeted NGS Fracti onati on and Capture hybridisation 1 - fractionation Figure from Mardis, E.R. (2012)

10 Targeted NGS Fracti onati on and Capture hybridisation 1 - fractionation 3 - enrichment Figure from Mardis, E.R. (2012)

11 Targeted NGS Fracti onati on and Capture hybridisation 1 - fractionation 3 - enrichment 4 - amplification Figure from Mardis, E.R. (2012)

12 Paired-end Reads 12

13 Paired-end Reads 13 Typically bp

14 Paired-end Reads 14 Typically bp nt nt ~50-100bp

15 Read-mapping 15 ~50-400nt linker ~50-400nt linker

16 Read-mapping 16 ~50-400nt linker

17 Read-mapping 17 ~50-400nt linker

18 Mapped WES reads viewed in IGV 18 Coverage Reads Exons

19 Variant Calling ideal 19

20 Variant Calling real world 20

21 Variant Calling Tools 21 SAMtools

22 Variant Calling Tools 22 SAMtools GATK

23 Variant Calling Tools 23 SAMtools GATK freebayes

24 Variant Calling Tools 24 SAMtools GATK freebayes Platypus

25 Variant Calling Tools 25 Variant calling tools will start by calling every potential variant they observe This will include true variants, and false-positives due to: Library preparation artefacts PCR artefacts Sequencing errors Mapping issues Algorithm issues Subsequently apply a number of mechanisms to attempt to help identify the true positives from the false-positives, and provide metrics Currently, you will always have some false positives, and some false negatives

26 Variant Calling Data 26

27 Annotation of variant call at sample level These fields are shown in the FORMAT field of the VCF, and there are values for every sample in a multi-sample VCF G A 999 PASS DP=180;VDB= e-01;RPB= e01;AF1=0.5;AC1=4; DP4=43,49,46,37;MQ=35;FQ=999;PV4=0.29,1,1,1 GT:PL:DP:SP:GQ 0/0:0,135,255:45:0:99 1/1:255,102,0:34:0:99 0/1:255,0,255:46:6:99 Tag Field Definition GQ Genotype Quality Phred-scaled confidence that the real genotype is that reported versus the next most likely genotype DP Depth PL Probability Likelihood The likelihood of the possible genotypes (order 0/0, 0/1, 1/1, 0/2, 1/2, 2/2 ), normalised such that the value for the reported GT is set to 0 SP Strand Bias p-value Phred-scaled strand bias p-value for sample Number of reliable base calls at this position

28 Annotation of variant call at pedigree level This information is taken from FILTER/INFO field of the VCF and indicate positions that are failing across samples Tag Field Definition sb0.05 / sb0.001 Strand Bias Indicates that there was a signficant bias towards variant calls only being observed on one strand across all samples at <0.05 or <0.001 respecively tdb0.05 Tail Distance Bias Indicates that there was a significant positional bias within reads for variant calls across all samples at this position mrd10 / mrd15 Minimum Read Depth Indicates that at least one of the samples had coverage <10 or <15 at this position msb30 Maximum Strand Bias Indicates that at least one of the samples had a strand bias (SP) >30 map Mappability Variant observed in a region to which we know we have problems to align short reads SALX=Y Samples At Least (covered) SAL10=3 would mean that at least 3 of the samples in the VCF have a read depth of 10 at this position

29 Indel identi fi cati on Raw BWA mapped reads 4 Following local realignment DePristo, M. et al. (2011)

30 Indel identi fi cati on Raw BWA mapped reads Following local realignment DePristo, M. et al. (2011)

31 Indel identi fi cati on where exactly? 31

32 Prioritize variants: advanced technical filtering Tail distance bias/read position bias No reads spanning this region ReadPosRankSum = ReadPosRankSum = ReadPosRankSum = samtools: VDB field (proportion), PV4 field (p-value) GATK: ReadPosRankSum field (Z-score)

33 Prioritize variants: advanced technical filtering Strand bias Reads Reads samtools: PV4 field (p-value) GATK: FS field (Phred-scaled p-value)

34 Raw Data Alignment Standardise Representation Illumina Platinum 50x WGS NA12878 NimbleGen MedExome 90x WES NA12878 BWA-MEM v0.7.8 GEM v3.1 Sort & Mark Duplicates (Picard) + Indel Realigment (GATK v3.3) Call Variants FreeBayes v0.9 Minimal Filtering QUAL >30 HaplotypeCaller v3.3 + GenotypeGVCF QUAL >30 SAMtools v1.2 1) fast (-Bug) 2) slow (-ug) QUAL >30 16 Final Call Set VCFs ( 8 Genome & 8 Exome)

35 NIST provide calls for callable regions of NA12878 genome i.e. excluding simplerepeats, CNVs and known segemental duplications, in this sample 2,191MB, 2,915,728 unique variant positions

36 Comparison with NIST set Intersect our VCFs to same regions and compared positions in our VCFs with the NIST VCF for concordance at level of Chr-Pos-Ref-Alt-GT Ignored positions that were multi-allelic for the alternative allele ~0.15% in NIST

37 Comparison with NIST - SNVs Dataset Total Calls TP FP FN Specificity Sensitivity F1 score Whole Genome SNVs NIST v2.18 Gold Standard BWA + FreeBayes BWA + HaplotypeCaller BWA + SAMtools fast BWA + SAMtools normal GEM3 + FreeBayes GEM3 + HaplotypeCaller GEM3 + SAMtools fast GEM3 + SAMtools normal TP Call identical in NIST and CNAG FP Call in CNAG not found in NIST FN Call in NIST not found in CNAG

38 Comparison with NIST - Deletions Dataset Total Calls TP FP FN Specificity Sensitivity F1 score Whole Genome Deletions NIST v2.18 Gold Standard BWA + FreeBayes BWA + HaplotypeCaller BWA + SAMtools fast BWA + SAMtools normal GEM3 + FreeBayes GEM3 + HaplotypeCaller GEM3 + SAMtools fast GEM3 + SAMtools normal TP Call identical in NIST and CNAG FP Call in CNAG not found in NIST FN Call in NIST not found in CNAG

39 Comparison with NIST - Insertions Dataset Total Calls TP FP FN Specificity Sensitivity F1 score Whole Genome Insertions NIST v2.18 Gold Standard BWA + FreeBayes BWA + HaplotypeCaller BWA + SAMtools fast BWA + SAMtools normal GEM3 + FreeBayes GEM3 + HaplotypeCaller GEM3 + SAMtools fast GEM3 + SAMtools normal TP Call identical in NIST and CNAG FP Call in CNAG not found in NIST FN Call in NIST not found in CNAG

40 Consensus between callers NIST callable region (2,195Mb)

41 Consensus between callers non-callable but mappable (594Mb)

42 WES V WGS

43 Summary Findings GEM3.1 is fast and resultant variant calling results are similar to those for BWA-MEM All variant callers tested are fairly similar in SNV accuracy There is much more variety in indel calls There is not a lot of diffence in accuracy between the two SamTools modes FreeBayes is very fast, but perhaps not as accurate as Haplotype Caller for indels

44

45

46 RD-Connect Genomics Platform D. Piscia, S. Laurie, S. Beltran

47 RD-Connect Genomics Platform Demos tomorrow at 14h and 16h D. Piscia, S. Laurie, S. Beltran

48 ISO 9001:2008

49 Acknowledgements rd-connect.eu platform.rd-connect.eu WP1: Coordination Hanns Lochmüller (Newcastle and TREAT-NMD) WP2: Patient registries Domenica Taruscio (ISS and EPIRARE) WP3: Biobanks Lucia Monaco (Fondaz. Telethon & EuroBioBank) WP4: Bioinformatics Christophe Béroud (INSERM Marseille) WP5: Unified platform Ivo Gut (CNAG Barcelona) WP6 Ethical/legal/social Mats Hansson (Uppsala) WP7: Impact/Innovation Kate Bushby (Newcastle and EUCERD/ EJARD) CNAG I. Gut S. Beltran D. Piscia S. Laurie J. Protasio A. Papakons. I. Martinez R. Tonda J.R. Trotta LUMC P.B. t Hoen M. Roos R. Raliyaperumal M. Thompson CNIO A. Valencia V. de la Torre J.M. Fernández A. Cañada U. Aveiro J.L. Oliveira P. Lopes P. Sernaleda Murdoch U. M. Bellgard MME H. Rehm AMU C. Béroud D. Salgado J.P. Desvignes Interactive BioSoftware A. Blavier S. Lair U. of Patras G. Patrinos Genesis S. Zuchner M.Gonzalez R. Acosta EGA D. Spalding J. Almeida-King A. Navarro J. Rambla Newcastle U. H. Lochmüller R. Thompson J. Dawson A. Topf I. Zaharieva U. Of Toronto M. Brudno M. Girdea S. Dumitriu O. Buske