TOWARDS ACCURATE GERMLINE AND SOMATIC INDEL DISCOVERY WITH MICRO-ASSEMBLY Giuseppe Narzisi, PhD Bioinformatics Scientist
July 29, 2014 Micro-Assembly Approach to detect INDELs 2 Outline 1 Detecting INDELs: issues and challenges 2 Scalpel micro-assembly pipeline 3 Applications 1 Large-scale validation experiment 2 De novo/transmitted mutations in Autism
July 29, 2014 Micro-Assembly Approach to detect INDELs 3 DETECTING INDELS Issues and challenges of variation discovery
July 29, 2014 Micro-Assembly Approach to detect INDELs 4 Why do we care? 1. Indels: insertion or deletion of DNA sequences. [The 1000 Genomes Project, Nature 2010] Second most common sources of variation in human genomes and the most common structural variant [Mullane et al. Hum. Mol. Genet. 2010] 2. Microsatellites: short tandem repeats (STRs), repeating sequences of 2-6 base pairs. Length changes are linked to more than 40 neurological diseases in humans. [Pearson et al., Nat. Rev. Gen. 2005] 3. De novo mutations: Autism: de novo mutations that are likely to severely disrupt the encoded protein are significantly more abundant in affected children than in unaffected siblings. [Ivan Iossifov et al., Neuron, 2012]
July 29, 2014 Micro-Assembly Approach to detect INDELs 5 The detection challenge Repeats Mapping errors Coverage Coverage 160 140 120 100 80 60 40 20 Father Mother Self Sibling 0 0 1000 2000 3000 4000 5000 6000 Genome location Irregularity in capture efficiency near the edges of the coding region SNP or Indel? Long insertion..ttgaattagccttggtgaattgagcctt...tttag--------agtgc..!! GAATGAGCC GAAT-GAGCC TTTAGAATAGGC! ATAGGCGAGTGC R SNP R
July 29, 2014 Micro-Assembly Approach to detect INDELs 6 Human Exome Capture technology: ~200,000 target regions 50-70 Mb (~2% of HG) 3-4 Gb 95% of exon-targets 400bp
July 29, 2014 Micro-Assembly Approach to detect INDELs 7 Localized Repeats in Human Exome Specificity challenge: 30% of exons have a perfect 10bp or larger repeat
July 29, 2014 Micro-Assembly Approach to detect INDELs 8 Tools sensitivity and performance Standard mapping and scanning algorithms (BWA, GATK, SAMTools, SOAP): 1 show low concordance on the same datasets. 2 suitable for detecting mutations only of a few nucleotides. SNPs INDELs Ti/Tv = ratio of the number of transition to transversion substitutions. When there is no bias there are twice as many possible transversions as transitions. Jason O'Rawe et al., Genome Medicine 2013, 5:28
July 29, 2014 Micro-Assembly Approach to detect INDELs 9 SCALPEL Micro-assembly pipeline
July 29, 2014 Micro-Assembly Approach to detect INDELs 10 Scalpel Novel DNA sequence micro-assembly pipeline to detect mutations within exome-capture data. Whole-Genome assembly Large scale genome structure Genotypic Heuristics to optimize resources (Time and Space) Micro-assembly Detect genome variations Haplotypic (Hom/Het state) Feasible to perform exhaustive search Features: 1. Self-tuning k-mer. 2. On-the-fly repeat composition analysis. 3. Exhaustive search of haplotypes. 4. Family pedigree: joint analysis of family members to detect de novo and transmitted mutations. NRXN1 de novo mutation
July 29, 2014 Micro-Assembly Approach to detect INDELs 11 Extract reads reference Build de Bruijn graph K = K+1 Remove low coverage nodes, dead-ends and compress Mark Source and Sink source yes If cycle or near-perfect repeat in any path source sink no sink Traverse graph and enumerate haplotype paths deletion insertion Align to reference
July 29, 2014 Micro-Assembly Approach to detect INDELs 12 Sliding window approach Extraction, assembly, alignment and INDEL detection performed in overlapping windows along the genome. 1. Localized assembly (smaller graph). 2. Minimize problem with coverage drops. 3. Distributed approach. 160 140 120 Father Mother Self Sibling Coverage 100 80 60 40 20 0 0 1000 2000 3000 4000 5000 6000 Genome location
July 29, 2014 Micro-Assembly Approach to detect INDELs Scalpel modes of operation Single exome - detects INDELs in one single dataset (e.g., one individual exome) De novo - detects de novo INDELs in one family of four individuals (mom, dad, aff, sib). Somatic: detects somatic INDELs in a tumor/sample pair given in input 13
July 29, 2014 Micro-Assembly Approach to detect INDELs 14 LARGE SCALE EXPERIMENT Re-sequencing of 1000 INDELs In collaboration with Lyon Lab
July 29, 2014 Micro-Assembly Approach to detect INDELs 15 Simulated Exome Sequencing ~200 thousand exons One INDEL per exon INDEL size: truncated lognormal distribution (μ=1,σ=10) in [1,100]. 100bp perfect reads Coverage distribution based on probe patterns (~30X). SAMtools, UnifiedGenotyper, FreeBayes and Platypus limited power to discover deletions larger than 40bp Scalpel, SOAPindel, and HaplotypeCaller perform the best.
July 29, 2014 Micro-Assembly Approach to detect INDELs 16 INDELs in one Exome Individual affected by Obsessive Compulsive Disorder (OCD), Tourette Syndrome, and Psychoses Captured using Agilent SureSelect v.2 and sequenced on the Illumina platform. 80% of the target at >20x coverage. 1000 INDELs for validation: 200 Scalpel 200 Haplotype Caller 200 SOAPindel 200 within intersection 200 long INDELs (>30bp) Hard to judge the quality of INDELs specific to each pipeline. Superior sensitivity or poor specificity??
July 29, 2014 Micro-Assembly Approach to detect INDELs 17 MiSeq validation PCR primers design: Primer 3 used to produce amplicons ranging in size from 200 to 350 bps centered around INDELs. Library construction: TruSeq DNA Sample Prep LS protocol from Illumina. MiSeq sequencing: 250 paired reads. avg cov: 10,000x count 15 10 5 0 15 10 5 0 15 10 5 0 15 10 5 0 15 10 HaplotypeCaller Intersected pipelines Large INDELs Scalpel Soap 100,000x 5 0 0 100000 200000 depth
July 29, 2014 Micro-Assembly Approach to detect INDELs 18 Focus on size distribution Bias towards deletions (for HaplotypeCaller 2.4.3) or insertion (for SOAPindel). Scalpel and HaplotypeCalle v3.0 instead show a well-balanced distribution between insertions and deletions
July 29, 2014 Micro-Assembly Approach to detect INDELs 19 Validated INDELs log total 10 Scalpel 77% PPV Invalid Valid 0-100 -80-60 -40-20 0 20 40 60 80 100 size log total 10 SOAPindel 50% PPV Invalid Valid 0-100 -80-60 -40-20 0 20 40 60 80 100 size log total 10 HaplotypeCaller 22% PPV Invalid Valid 0-100 -80-60 -40-20 0 20 40 60 80 100 size INDELs not passing validation correlate well with size bias.
July 29, 2014 Micro-Assembly Approach to detect INDELs 20 Validated INDELs log total 10 Scalpel 77% PPV Invalid Valid 0-100 -80-60 -40-20 0 20 40 60 80 100 size log total 10 SOAPindel 50% PPV Invalid Valid 0-100 -80-60 -40-20 0 20 40 60 80 100 size log total 10 HaplotypeCaller 22% PPV Invalid Valid 0-100 -80-60 -40-20 0 20 40 60 80 100 size INDELs not passing validation correlate well with size bias.
July 29, 2014 Micro-Assembly Approach to detect INDELs 21 HaplotypeCaller GATK 2.8 Additional validation experiments Frequency 1000 100 10 1 GATK v2.4-3 -100-80 -60-40 -20 0 20 40 60 80 100 Size (base pairs) Raw Hard-filter Frequency 1000 100 10 1 GATK v2.8-100 -80-60 -40-20 0 20 40 60 80 100 Size (base pairs) Frequency 1000 100 10 1 GATK v3.0-100 -80-60 -40-20 0 20 40 60 80 100 Size (base pairs) Scalpel shows a significantly higher validation rate for longer indels (size >5bp)
July 29, 2014 Micro-Assembly Approach to detect INDELs 22 Microsatellites instability Position based match Exact match (position+sequence) Due to instability and higher error rate (e.g., homopolymers): Multiple candidates at microsatellite loci. Ref:..TAGCC- AAAAAAATGGTGC..! Alt:..TAGCC-AAAAAAAATGGTGC.. Alt:..TAGCCAAAAAAAAATGGTGC.. Different tools report different signatures.
July 29, 2014 Micro-Assembly Approach to detect INDELs 23 Characterization of False Discovery Rate (1) Outside of microsatellites Within microsatellites 614 indels detected by Scalpel and validated by re-sequencing. Higher FDR for mutations within microsatellites. Recommendation on how to select a chi-square score cutoff to achieve a given FDR.
July 29, 2014 Micro-Assembly Approach to detect INDELs 24 Characterization of False Discovery Rate (2) 250 200 Valid Invalid Coverage 150 100 50 0 0 2 4 6 8 10 12 Chi-Square Score Mutations that do not pass validation (red dots) are generally characterized by lower coverage.
July 29, 2014 Micro-Assembly Approach to detect INDELs 25 POPULATION SCALE INDELS ANALYSIS Simons Simplex Collection In collaboration with Wigler Lab
July 29, 2014 Micro-Assembly Approach to detect INDELs 26 Simons Simplex Collection ~2700 families. Quad: two parents, one affected child and one unaffected child. NimbleGen SeqCap EZ Exome v2.0 (36 Mb). Illumina HiSeq: ~93bp reads after removing barcodes. Three major studies reporting strong enrichment for de novo gene desrupting mutations in autistic kids: 1 CSHL: Iossifov et al. (2012) Neuron. 74:2 285-29 2 Yale: Sanders et al. (2012) Nature. 485, 237 241. 3 WashU: O Roak et al. (2012) Nature. 485, 246 250.
July 29, 2014 Micro-Assembly Approach to detect INDELs 27 Transmitted INDELs 1303 families (5212 individuals) Database with > 12 million INDELs Increased power to detect insertions. Subdivide by DNA context. Goal: discover significant biology that was impossible to measure a few year ago.
July 29, 2014 Micro-Assembly Approach to detect INDELs 28 INDEL counts
July 29, 2014 Micro-Assembly Approach to detect INDELs 29 Frame-preserving INDELs are more abundant within coding sequences (CDS)
July 29, 2014 Micro-Assembly Approach to detect INDELs 30 Deletion-to-Insertion ratio! Category Insertions Deletions DI-ratio Total Intron 6295 11490 1.82 17785 Intergenic 310 645 2.08 955 Frame-shift 1207 2758 2.28 3965 Non-frame-shift 622 2049 3.2 2671 UTR 790 1377 1.74 2167 Splice site 56 196 3.5 252 All 9280 18515 1.99 27795 Twice as many deletion than insertions across all annotation category. In agreement with other recent studies: Montgomery et al., Genome Res. Published in Advance March 11, 2013. Sjodin et al. Plos ONE, 2010, Volume 5, Issue 1.
July 29, 2014 Micro-Assembly Approach to detect INDELs 31 Transmitted Frame-shifts at lower frequency In agreement with MacArthur et al., 2010. Loss-of-function variants in the genomes of healthy humans. Hum. Mol. Genet. 19 (R2): R125-R130
July 29, 2014 Micro-Assembly Approach to detect INDELs 32 DE NOVO MUTATIONS IN AUTISM Simons Simplex Collection In collaboration with Wigler Lab
July 29, 2014 Micro-Assembly Approach to detect INDELs 33 Unified Genetic Model of Autism Sporadic Autism Sporadic autism: spontaneous mutation with high penetrance in males and relatively poor penetrance in females Family autism: offspring, most often females, who carry a new causative mutation but are unaffected and in turn transmit the mutation in dominant fashion to their offspring. Familial Autism Legend Sporadic muta-on Fails to procreate Michael Wigler et al. PNAS 2007
July 29, 2014 Micro-Assembly Approach to detect INDELs 34 Risk by kind of rare genetic variants Population-level enrichment in ASD versus controls of each form of rare mutation studied to date INDELs? Neuron, 2013, volume 77, Issue 2, 23
July 29, 2014 Micro-Assembly Approach to detect INDELs 35 De novo INDELs in Autism 1303 families: 686 CSHL, 308 State Lab, and 309 Eichler Lab # INDEL&effect& Aut& Sib& Aut&M& Aut&F& Sib&M& Sib&F& Total& Frame&shift& 85# 47# 66# 19# 31# 16# 132# Intron& 29# 32# 23# 6# 13# 19# 61# Intergenic& 4# 0# 4# 0# 0# 0# 4# No&frame&shift& 9# 9# 9# 0# 3# 6# 18# Splice=site& 3# 1# 3# 0# 1# 0# 4# UTR& 5# 2# 5# 0# 0# 2# 7# Total& 135# 91# 110# 25# 48# 43# 226# De novo INDELs that are likely to severely disrupt the encoded protein are significantly more abundant in affected children than in unaffected siblings Significant overlap (25 out of 132) between the LGD target genes and the set of 842 FMRP-associated genes (Darnell, J.C. et al. Cell, 146:247 261, 2011).
July 29, 2014 Micro-Assembly Approach to detect INDELs 36 De novo filters 1 Population filter: Eliminate candidate loci that are common in the population (focus on ultra rare events). 2 Re-assembly each region around candidate INDELs using more sensitive parameters (max sensitivity): Reduce starting k-mer value to 10. turn off removal of low coverage nodes. 3 Final filters: Parents coverage 15x. chi-square (Χ 2 ) coverage score 10.84 (corresponding to a p-value of 0.001). Χ! =!!!!!!!!!!!!!! +!!!!!!!
July 29, 2014 Micro-Assembly Approach to detect INDELs 37 Some statistics Most de novo mutations are small [1bp-6bp] and could be found by the GATK-based pipeline. For a few loci, the sizes of deletions were corrected by micro-assembly compared to BWA+GATK protocol. Validation rate re-sequencing of 102 candidate INDELs; 84 were confirmed as de novo mutations, 11 were invalid and 7 failed, giving an 82% positive predictive rate for Scalpel.
July 29, 2014 Micro-Assembly Approach to detect INDELs 38 Suspicious deletion aussc12252, chr16:1412556 Suspicious set of reads starting at the same position
July 29, 2014 Micro-Assembly Approach to detect INDELs 39 Suspicious deletion Soft-clipped bases Suspicious high number of soft-clipped bases
July 29, 2014 Micro-Assembly Approach to detect INDELs 40 Large 33bp de novo deletion MiSeq validation 33bp deletion!
July 29, 2014 Micro-Assembly Approach to detect INDELs 41 IGV snapshot aussc13578, 13:25280526 1bp del 5bp del
July 29, 2014 Micro-Assembly Approach to detect INDELs 42 Complex de novo event Deletion of 6 bp together with two adjacent SNPs detected in only one of the two haplotypes of the autistic child (aussc13578, 13:25280526). Detected as two different deletions (1bp and 5bp) by the GATK-based sequencing pipeline.
July 29, 2014 Micro-Assembly Approach to detect INDELs 43 CONCLUSION
July 29, 2014 Micro-Assembly Approach to detect INDELs 44 Conclusion Assembly is the missing link towards high accuracy and increased power for INDEL mutation discovery: Allows the algorithm to break free from the expectations of the reference. Extended power crucial for analysis of inherited and somatic mutations. Scalpel: highly accurate tool to detect de novo, transmitted, and somatic INDELs. Errors of current detection software explained by a largescale re-sequencing experiment. Population wide analysis: de novo INDELs in Autism.
July 29, 2014 Micro-Assembly Approach to detect INDELs 45 Software & reference Open source project at: http://scalpel.sourceforge.net/ Paper preprint: Narzisi G., O'Rawe J.A., Iossifov I., Lee Y., Wang Z., Wu Y., Lyon G.J., Wigler M., Schatz M.C. Accurate detection of de novo and transmitted INDELs within exome-capture data using micro-assembly. CSHL BioRxiv (DOI: 10.1101/001370). Manuscript accepted for publication in Nature Methods.
July 29, 2014 Micro-Assembly Approach to detect INDELs 46 Acknowledgment Michael C. Schatz Michael Wigler Gholson J. Lyon Ivan Iossifov ADHD project Jason O Rawe Han Fang Yiyang Wu Autism project Dan Levy Michael Ronemus Yoonha Lee Zihua Wang Ewa Grabowska Peter Andrews Mitchell Bekritsky Jude Kendall
July 29, 2014 Micro-Assembly Approach to detect INDELs 47 THANK YOU Email: gnarzisi@nygenome.org