Slide 1: DRAGEN v4.3: Comprehensive coverage with leading edge innovations
M-US-00466 DRAGENTM v4.3: Comprehensive coverage with leading edge innovations Yi Lian Shyamal Mehtalia Staff Product Manager, Illumina Principle Software Engineer, Illumina For Research Use Only. Not for use in diagnostic procedures.
Slide 2: DRAGEN is a collection of bioinformatics pipelines for secondary analysis
DRAGEN is a collection of bioinformatics pipelines for secondary analysis Library Prep Sequencing Secondary Tertiary Analysis Analysis and Reporting DRAGEN (Dynamic Read Analysis for GENomics) Accurate, comprehensive, and efficient secondary analysis. For Research Use Only. Not for use in diagnostic
Slide 3: Award winning accuracy for germline and somatic variant calling
Award winning accuracy for germline and somatic variant calling Germline Somatic PrecisionFDA Truth Challenge V2 PrecisionFDA NCTR Indel Calling from Oncopanel Sequencing Data Announced August 2020 Announced September 2022 DRAGEN team won best performance in Difficult-to-Map Regions and All Benchmar
Slide 4: The right solution for your needs
The right solution for your needs Updated for DRAGEN v4.3 High Medium DRAGEN on-instrument DRAGEN on DRAGEN on Illumina DRAGEN on-premise DRAGEN Multi-Cloud (NextSeqTM 1000/2000, BaseSpaceTM Sequence Connected AnalyticsTM server Bring Your Own License NovaSeqTM X Series) Hub (BSSH) (ICA) (AWS, Azure
Slide 5: DRAGEN v4.3
DRAGEN v4.3 A comprehensive genome with industry-leading innovations More accurate Comprehensive Enhanced efficiency than ever coverage & accessibility Next generation multigenome New specialized callers ORA Compression (graph) with 128 samples Variant Calling in segmental Expanded support for human
Slide 6: DRAGEN Germline
DRAGEN Germline For Research Use Only. Not for use in diagnostic procedures.
Slide 7: Accuracy
DRAGEN Secondary Analysis Most accurate secondary analysis in all-benchmark regions* DRAGEN Other secondary F1, All Benchmark Regions on Illumina Data analysis platforms 100.00% • Available since v4.0, continuous 99.89% improvements of DRAGEN Machine 99.84% 99.83% 99.80% Learning (ML) detects varian
Slide 8: DRAGEN continues to set new standards for accuracy
DRAGEN continues to set new standards for accuracy Change in FN+FP, SNPs+INDELs HG002 NISTv4.2.1 FN FP 45,000 83% Error Reduction 39,046 40,000 35,000 12,135 30,000 25,000 FP + FN 21,479 20,402 40% Error 20,000 Reduction Next-Gen 7,812 7,278 multigenome 13,793 15,000 12,851 12,815 26,911 11,163 3,36
Slide 9: The most diverse set of data built-in in the next-gen multigenome reference
The most diverse set of data built-in in the next-gen multigenome reference Enabling improved variant calling accuracy across populations Setting standards for population representation across the European Ancestry globe 28 (22%) African Ancestry 42 (33%) Extends sample population to 128 samples of
Slide 10: Germline small variant caller – SNP accuracy
Germline small variant caller – SNP accuracy Reduce SNV FP+FN by 35% on average cross population Change in FP+FN, WGS SNPs, NIST v4.2.1 12,000 10,310 10,000 9,103 8,879 9,031 2,488 8,412 8,289 8,000 1,837 2,120 2,114 6,572 6,520 2,587 2,245 FP + FN 5,694 5,706 5,682 5,973 6,000 4,934 2,357 FP 2,650
Slide 11: Germline small variant caller - INDEL accuracy
Germline small variant caller - INDEL accuracy Reduce indel FP+FN by 12.1% on average cross population Change in FP+FN, WGS indels, NIST v4.2.1 3,000 2,631 2,500 2,387 2,060 2,102 1,078 1,949 1,920 2,000 1,829 1,795 1,050 1,681 1,614 FP + FN 716 806 1,569 1,500 761 1,380 1,358 857 617 666 792 FP 1,1
Slide 12: Personalized Reference (beta) for small variant calling
Personalized Reference (beta) for small variant calling DRAGEN builds a 2-haplotypes personalized reference to impute variants, use as priors in the Variant Caller, create new personalized ML model # To enable personalized reference: --enable-personalization=true (default to false) Reduces FP+FN by
Slide 13: Personalized Germline Small Variant Caller – Accuracy
Personalized Germline Small Variant Caller – Accuracy Personalization reduces FP+FN by 28.7% for SNPs, and 7.9% for indels compared to 4.3 * Change in FP+FN, WGS SNPs + indels, NIST v4.2.1 14,000 FP FN 12,412 12,000 11,163 10,828 10,826 False negatives and false positives 3,294 9,792 9,858 10,000 2,
Slide 14: DRAGEN provides tools to build custom multigenome references
DRAGEN provides tools to build custom multigenome references Build reference genome tailored to your population Customers can build high-quality custom multigenome references from pangenome data Custom Reduces ethnicity bias and allows customization Assemblies of multigenome references for populatio
Slide 15: Improve small variant calling accuracy with custom references
Improve small variant calling accuracy with custom references DRAGEN HPRC-based multigenome reference yields better accuracy results than DRAGEN v4.2 Linear reference and HPRC 44-samples graph. Change in FP+FN, WGS SNPs + indels, NIST v4.2.1 FP FN 30,000 26,598 False negatives and false positives 24
Slide 16: Introducing mosaic variant caller
Introducing mosaic variant caller Mosaicism is a postzygotic mutation that leads to some cells in the body having different DNA than others Mosaic variants occur at low allele frequencies, making them hard to detect To study the effects of mosaicism on biology and disease, mosaic variants need to be
Slide 17: Detect Mosaic variants at low allele frequency
Detect Mosaic variants at low allele frequency Enable Mosaic variant detection at low allele frequency New in v4.3 - integrated mosaic caller Map/Align FASTQ Reference Integrated single workflow with germline variant Or DRAGEN-VC DRAGEN-ML runs automatically with calling. Facilitated by Machine Lear
Slide 18: Greatly Improved Recall for Low Allele Frequency Variants
Greatly Improved Recall for Low Allele Frequency Variants Higher sensitivity for variant calls at low AF, with low added FPs Change in FP+FN, WGS, HG002 (NIST v4.2.1) + spike FN spiked variants VAF stratification 25000 21715 1857 20000 15792 7293 15000 13468 1610 FP + FN False negatives 3372 5387 10
Slide 19: DRAGEN is more accurate than other mosaic callers
DRAGEN is more accurate than other mosaic callers End-to-end analysis with fast turnaround time Mosaic variant calling accuracy comparison Mosaic variant calling runtime comparison Runtime (hr) Hardware DRAGEN Server DRAGEN v4.3 0.3 v4 Multi-core CPU + DeepMosaic 5.5 NVIDIA A100 GPU MosaicForecast 1
Slide 20: Greatly improved precision for high-depth samples with fast turnaround time
Greatly improved precision for high-depth samples with fast turnaround time Change in FP+FN, WGS SNPs + indels, NIST Mosaic v0.1 3,500 73% 3,000 False negatives and false positives 2,500 2,000 FP 3249 1,500 FN Limit of Detection at 2.67% AF for SNPs and 8.8% AF for Indels on 1100x WES admixtures*. 1
Slide 21: DRAGEN provides accurate genotyping of clinically relevant genes in segmental duplications
DRAGEN provides accurate genotyping of clinically relevant genes in segmental duplications Segmental duplications (SegDup) are 1 kb+ regions with ≥ 90% similarity Copy 1 Copy 2 Contain sequences that are relevant to X X X X X X X X X X X X disease Represent 5% of the genome, and have poor mappabilit
Slide 22: Multi-region joint detection (MRJD) enables de novo germline small variant calling in paralogous regions
Multi-region joint detection (MRJD) enables de novo germline small variant calling in paralogous regions DRAGEN MRJD Caller (NEW!) Gene Application | Conditions Hereditary Cancer Screening PMS2 | e.g. Lynch Syndrome for ✓ Haplotype-based variant calling from collected reads Colorectal/Endometrial
Slide 23: Multi-region joint detection (MRJD) enables de novo germline small variant calling in paralogous regions
Multi-region joint detection (MRJD) enables de novo germline small variant calling in paralogous regions Report uniquely placed and 1 Build and place all haplotypes using all reads 2 region-ambiguous variants For Research Use Only. Not for use in diagnostic procedures. 23
Slide 24: MRJD empowers researches on clinically relevant and challenging genes
MRJD empowers researches on clinically relevant and challenging genes • Benchmarked on more than 100 cell lines per gene (using RTG tools squash ploidy mode) • MRJD High Sensitivity mode offers high recall at the expense of precision, with low spurious call rate For Research Use Only. Not for use in
Slide 25: Detect copy number variations in segmental duplications
Detect copy number variations in segmental duplications Example Command Line: DRAGEN v4.3 rescues ~1Mbp of CNV bins previously dragen \ -r <HASHTABLE> \ excluded from analysis. --output-directory <OUTPUT> \ --output-file-prefix <SAMPLE> \ --ref-dir <HASHTABLE> \ • Improves CNV detection across 43 cl
Slide 26: CNV bin rescuing allows for more accurate CNV calls in segmental duplication regions
CNV bin rescuing allows for more accurate CNV calls in segmental duplication regions E perimentally confirmed C P2A C P2A gene fusion (HG01 ) identified by rescuing previously e cluded bins REF DE REF Comparison against experimentally validated CNV calls in segdup genes No. Correct by Method & sampl
Slide 27: Enhanced CNV calling using WES data
Enhanced CNV calling using WES data New filters reduce false positive rates • Likelihood Ratio (FORMAT: LR) – Log10 likelihood ratio Depth estimation of ALT to REF 2 1.5 Likelihood ratio • Dinucleotide Biases (FORMAT: GC/CT/AC) – Measure of 1 dinucleotide biases that are outside of typical ranges wi
Slide 28: Improved structural variant calling
Improved structural variant calling Method MEI Recall SV accuracy improvements DRAGEN 4.3 0.885 Robust assembly methods in repeat regions improves DRAGEN 4.2 0.867 insertion recall1 Mobster 0.2.4.1 0.756 DRAGEN SV retains higher recall in Mobile Element Insertions (MEI) when compared to dedicated me
Slide 29: Improve structural variant recall with new VNTR caller
Improve structural variant recall with new VNTR caller SV Accuracy • Variable Number Tandem Repeats (VNTRs) – 100% tandem repeat (TR) with pattern size ≥10 bp, larger 80% +8.5% than STRs. 60% +11% • New VNTR caller for large variant calls in TRs 40% 20% • Supports hg38, hg19, GRCh37 references with
Slide 30: VNTR Caller - Detects expansions and contractions in tandem repeat regions
VNTR Caller - Detects expansions and contractions in tandem repeat regions DRAGEN-VNTR reports calls following the VCFv4.4 spec: chr1 2141862 vntr_00000216 A • Enabled with option “--enable-vntr true” <CNV:TR>,<CNV:TR> 0.946 PASS SVTYPE=CNV;EVENTTYPE=VNTR,VNTR;SVLEN=634,634;END=2142 • Reports calls
Slide 31: Discover new STR expansions with the STR profiler module
Discover new STR expansions with the STR profiler module Cases vs controls bonf contig start end motif pvalue counts ExpansionHunter de novo is now in DRAGEN with faster pvalue compute times and improved ease of use chr1 1246 1347 AAAG 0.3356 1 […] • Compute sample profiles while mapping: Outlier •
Slide 32: STR expansions are rediscovered in affected individuals
STR e pansions are “rediscovered” in affected individuals Dolzhenko et al., 2020 For Research Use Only. Not for use in diagnostic procedures. 32
Slide 33: STR profiling module – streamlining the workflow
STR profiling module – streamlining the workflow For Research Use Only. Not for use in diagnostic procedures. 33
Slide 34: Unified VCF output for three targeted callers
Unified VCF output for three targeted callers Improves efficiency for downstream interpretations • Enabled for HBA, GBA and CYP21A2 X Z HBA2 X Z HBA1 • Records with ambiguously placed variants can be One-gene deletion (-α3.7) One-gene deletion (-α4.2) PASS even with low QUAL and low GQ values Two-ge
Slide 35: HBA Structural Variant VCF Output
HBA Structural Variant VCF Output Targeted calling for these 3 genes • Forced genotyping of all 5 supported structural variants: is enabled by default when small • -α3.7 DEL and ααα3.7 DUP VC is enabled (only for WGS). • -α4.2 DEL and ααα4.2 DUP • --SEA DEL • ALLELE_ID INFO field used to label varia
Slide 36: VCF Output for Region-Ambiguous Small Variants
VCF Output for Region-Ambiguous Small Variants • Enabled for HBA, GBA and CYP21A2 { • Records can be PASS even with low QUAL and low GQ values "sampleId": "WGS1000-P4-A5", • These variants require further assay if placement "cyp21a2": { in target gene is required "totalCopyNumber": 4, • VCF format c
Slide 37: Small Variant VCF Output for Recombinant Haplotypes
Small Variant VCF Output for Recombinant Haplotypes • Enabled for GBA and CYP21A2 { • Gene conversions reported as small variants in "sampleId": "HG00422", target gene rather than SV breakends "gba": { • Deletion-like recombinant haplotypes reported as small variants in target gene with overlapping
Slide 38: DRAGEN runtime updates
DRAGEN runtime updates More accurate, more comprehensive, with small runtime increase DRAGEN Germline Runtime* 40 Approx. 15-20% runtime increase 34.7 35 compared to v4.2 30 29.5 Includes mapping with multigenome 26 Runtime (min) 25 reference and ML calibration 21 20 •26 minutes*- Map, align, and sm
Slide 39: Cohort analysis using gVCF Genotyper
Cohort analysis using gVCF Genotyper For Research Use Only. Not for use in diagnostic procedures.
Slide 40: Improved multi-sample VCF output
Improved multi-sample VCF output To enable sparse msVCF output : New options to compress and customize multi- sample VCF (msVCF) output --gg-output-type=spVCF --gg-squeeze-msvcf=true Sparse compression of msVCF output (Michael Lin, 2020), yields 20-30x size reduction for large How to customize msVCF
Slide 41: New msVCF metrics in gVCF Genotyper
New msVCF metrics in gVCF Genotyper New metrics enable filtering for highly accurate call sets Allelic balance can identify incorrect Genotype calls To enable output of allelic balance in the msVCF: --gg-msvcf-info-fields=ABHom;ABHet;ABHetP To filter msVCF based on maximum P-value of allelic balance
Slide 42: Performance improvements enable Biobank analysis
Performance improvements enable Biobank analysis New gVCF Genotyper callset released for the UK Biobank Improved msVCF generation in DRAGEN 4.3 • 500k UKBB WGS samples were re-processed with DRAGEN ML recalibration and gVCF Wallclock runtime RAM usage Genotyper on ICA • Aggregated 1.4 billion varian
Slide 43: DRAGEN Somatic
DRAGEN Somatic For Research Use Only. Not for use in diagnostic procedures.
Slide 44: General somatic VC improvements
General somatic VC improvements Sample variant rescued by columnwise detection: • Repetitive local sequence • Columnwise detection is now enabled by leads to cycles in graph default, improving small variant recalls in assembly of reads and repetitive regions may result in missed variants • Columnwis
Slide 45: Allele specific systematic noise files for somatic variant calling
Allele specific systematic noise files for somatic variant calling • New v2 noise file format encodes mean and max noise levels • Allele-specific filtering • Generate custom systematic noise files ➢ Run normals using tumor-only workflow in sensitive mode --vc-detect-systematic-noise # optional --vc-
Slide 46: ASCN calling in somatic WES analysis
ASCN calling in somatic WES analysis Extends ASCN detection in tumor-only mode Example DRAGEN command: --tumor-bam-input=<TUMOR_BAM> Allele-specific copy number (ASCN) analysis extends the --cnv-population-b-allele-vcf=<CNV_POP_VCF> --enable-cnv=true utility of copy number alteration by enabling the
Slide 47: New DUX4 caller in somatic WGS analysis
New DUX4 caller in somatic WGS analysis DUX4 gene related rearrangements detection DUX4 gene (DUX4-r) rearrangements are involved in a To enable DUX4 caller : --enable-dux4-caller=true subtype of acute lymphoblastic leukemia (ALL) • DUX4 rearrangement detection enabled by machine Example output from
Slide 48: DUX4 caller
DUX4 caller DUX4 gene related rearrangements detection from WGS data Samples Core DUX4 region 100 kb upstream of DUX4L28 Extended DUX4 region End of sequenced region • Define customized DUX4 calling regions IGH (chr14) DUX4 (chr4/chr10) Paired spanning read • Generate hypothesis for somatic-only WGS
Slide 49: DRAGEN RNA
DRAGEN RNA For Research Use Only. Not for use in diagnostic procedures.
Slide 50: Improved RNA fusion detection by updated machine learning
Improved RNA fusion detection by updated machine learning Extensively trained machine learning model improves RNA fusion detection accuracy Attribute Old Model (pre-4.3) New Model (4.3) Model Linear Non-linear Updated non-linear model with more features Features Used 8 39 Significantly enhanced trut
Slide 51: Improved RNA fusion detection by updated machine learning
Improved RNA fusion detection by updated machine learning Extensively trained machine learning model improves RNA fusion detection accuracy • Updated non-linear model with more features • Significantly enhanced truth set, using simulations + real data • Additional datasets cross library preps (WTS,
Slide 52: New DRAGEN RNA Splice Variant Caller (beta)
New DRAGEN RNA Splice Variant Caller (beta) Integrates the caller used in DRAGEN TSO500 Enables detection of alternative splicing or intra- genic fusions Targets well-known splice variant biomarker events*: • EGFR VIII • MET exon14 • AR-V General (non-targeted) option available with Z increased fals
Slide 53: RNA Splice Variant Caller Resource Files
RNA Splice Variant Caller Resource Files Outputs Required Resource Files • Anchored-RNA reference HT (for mapper) • Gene Annotation file (i.e. GTF or GFF) Optional (recommended) New Resource Files • rna-splice-variant-knowns: Knowns SJ file (“whitelist”) <prefix>.splice_variants.csv • rna-splice-var
Slide 54: ORA Compression
ORA Compression For Research Use Only. Not for use in diagnostic procedures.
Slide 55: Compress non-human data at high compression ratio
Compress non-human data at high compression ratio List of supported species : ✓ Homo sapiens – bisulfite (methylated data) ✓ Glycine max (soybean) ✓ Sus scrofa (pig) ✓ Rattus norvegicus (Norway rat) ✓ Gallus gallus (chicken) ✓ Zea mays (maize) ✓ Oryza sativa (Japanese rice) ✓ Danio rerio (zebrafish)
Slide 56: ORA Compression supports more datatypes and species
ORA Compression supports more datatypes and species Reduces file size and storage cost by up to 75% Compression improvement on human bisulfite data Compression improvement on pig data dataset: B-Hela_S2 WGS 45x sequenced on NovaSeq6K, compression on an dataset: SRR14901873 WGS 25x sequenced on NovaS
Slide 57: DRAGEN Imputation & Population Haplotyping
DRAGEN Imputation & Population Haplotyping For Research Use Only. Not for use in diagnostic procedures.
Slide 58: Boost small variant calling accuracy from low-pass data
Boost small variant calling accuracy from low-pass data • New reference panel (IRPv2.1, human hg38) improves IRPv2.1 IRPv2.0 indel imputation accuracy Data source1 3,202 samples 3,202 samples • Supports up to 100 input fastq samples or 1000 input Multi-allelic SNP Yes Yes positions vcf samples1 INDE
Slide 59: Greatly improved indel imputation accuracy
Greatly improved indel imputation accuracy Imputation performance on 1x HG002 with IRPv2.0 and IRPv2.1 Sensitivity SNPs INDELs Sensitivity ~14,000 less FN FP FP IRPv2.0 IRPv2.1 For Research Use Only. Not for use in diagnostic procedures.
Slide 60: DRAGEN Imputation and Population Haplotyping
DRAGEN Imputation and Population Haplotyping Handle batch of samples through an end-to-end workflow on BSSH and ICA Imputation pipeline (BSSH/ICA) Population Haplotyping pipeline (ICA) infers variants of low-pass sequencing data builds haplotypes from a set of population samples (Integration of Shap
Slide 61: DRAGEN Imputation
DRAGEN Imputation Higher gain of accuracy for coverage between 0.5-4x SNPs INDELs SNPs 250,000 120,000 300,000 200,000 100,000 250,000 Count of FP+FN Count of FP+FN Count of FP+FN 80,000 200,000 150,000 159,763 76,895 60,000 70,223 150,000 115,369 64,017 100,000 40,000 100,000 74,090 50,000 20,000 7
Slide 62: DRAGEN Population Haplotyping
DRAGEN Population Haplotyping End-to-end Population Haplotyping pipeline on ICA INPUTS Population Haplotyping workflow OUTPUTS Unphased msVCF Pre-processing on single chromosome1 and Phase Phase Rare Common multi-nodes multi-nodes Genetic map and config file (.tar.gz) Phased msVCF on single chromoso
Slide 63: DRAGEN Population haplotyping
DRAGEN Population haplotyping Chunking size impacts accuracy, runtime, and cost – ICA workflow includes optimized settings ICA runtime1 and Switch Error Rate (SER) with different Imputation accuracy with IRPv2.0 built with different chunking sizes to build IRPv2.0 on Chr1 chunking sizes on HG002 1x
Slide 64: Fractional Downsampler
Fractional Downsampler For Research Use Only. Not for use in diagnostic procedures.
Slide 65: New DRAGEN Fractional Downsampler Method
New DRAGEN Fractional Downsampler Method Downsample high coverage samples to simulate different amounts of sequencing DRAGEN input DRAGEN Fractional Map/Align Downstream callers Downsampler All input formats supported Downsampling rate is provided by the by DRAGEN can be used user and is applied to
Slide 66: New Fractional Downsampler
New Fractional Downsampler Simulates different amount of sequencing for high coverage samples To enable the fractional downsampler: • Fully integrated workflow with downstream analysis --enable-fractional-down-sampler=true • Subsampling based on user-defined percentage of reads To set percentages of
Slide 67: Illumina Connected Annotations (Nirvana)
Illumina Connected Annotations (Nirvana) For Research Use Only. Not for use in diagnostic procedures.
Slide 68: Connected Annotations
Connected Annotations • Build on the foundation of a previous tool called Nirvana An Illumina Genome annotated with PrimateAI-3D score Illumina Connected annotations is the latest version of the can reduce VUS and predict pathogenicity of all protein annotation engine, incorporating PrimateAI-3D as
Slide 69: PrimateAI-3D utilizes primate variants and protein structures to solve VUS in human genome coding regions
PrimateAI-3D utilizes primate variants and protein structures to solve VUS in human genome coding regions Our current knowledge of the clinical PrimateAI-3D, A deep learning model trained effects of genetic variants is nascent on millions of benign primate variants Fraction of Variants Known (0.1%)
Slide 70: An Illumina Genome annotated with PrimateAI-3D score can reduce VUS
An Illumina Genome annotated with PrimateAI-3D score can reduce VUS and predict pathogenicity of all protein coding variants Benefits customers in genetic disease and oncology researches PrimateAI-3D reclassified >4 million human missense Fraction of Variants variants of previously unknown consequen
Slide 71: PrimateAI-3D improves genetic risk prediction using rare variants
PrimateAI-3D improves genetic risk prediction using rare variants PrimateAI-3D PRS is highly accurate in both European and non-European populations PrimateAI-3D achieves top performance across PrimateAI-3D accurately predicts phenotypes in six benchmarks for variant interpretation 450,000 individual
Slide 72: Thank you!
Thank you! Illumina’
Slide 73: Appendix
Appendix Illumina’
Slide 74: New DRAGEN Report tool
New DRAGEN Report tool Generate comprehensive, interactive HTML reports from DRAGEN's output files Flexible deployment options with standalone docker image Generate reports without re-running a whole workflow Mix and match samples from across workflows into one report Supports current features and n
Slide 75: Coming soon – multi-version management on DRAGEN Server
Coming soon – multi-version management on DRAGEN Server Current – one DRAGEN version on one Future – multiple versions on each DRAGEN server at a time DRAGEN Server V4.3 V4.3 Q3 2024 v4.1.23 v4.1.23 v2.1 v2.1 TSO 500 TSO 500 • New DRAGEN installer allows simultaneous hosting of DRAGEN software v4.3