> For the complete documentation index, see [llms.txt](https://help.connected.illumina.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.connected.illumina.com/dragen/dragen-v4.6/reference/release-notes-readme/dragen-crn-v4.6.2.md). # DRAGEN v4.6.2 Release Notes *** ## Introduction These release notes detail the key changes to software components for the Illumina® DRAGEN™ Secondary Analysis Software v4.6. Changes are relative to DRAGEN™ v4.5. If you are upgrading from a version prior to DRAGEN™ v4.5, please review the release notes for a list of features and bug fixes introduced in subsequent versions. DRAGEN™ Installers, Resource Files, and Release Notes are available here: DRAGEN™ User Guide is available here: The software package includes downloadable installers for Phase 3 and Phase 4 on-premises servers: * DRAGEN™ SW for x86 Oracle 8 — `dragen-4.6.2-12.multi.el8.x86_64.run` * DRAGEN™ SW for x86 Oracle 9 — `dragen-4.6.2-12.multi.el9.x86_64.run` The following configurations containing DRAGEN™ 4.6 are also available on request: * el8/9 Amazon Machine Images (AMIs) for f2 instances * AlmaLinux 8 Microsoft Azure Image (VM) * el8/9 compatible RPM packages for use with Amazon Web Services (AWS) f2 instances * DRAGEN™ Kernel drivers for el8/9 * **Software Mode** installer (`dragen-softwaremode-*.bin`) for x86 el8/9 — see [DRAGEN Software Mode](#dragen-software-mode) DRAGEN™ v4.6 is also made available on: * Illumina BaseSpace and Platform Core * AWS and Azure Marketplaces *** ## Contents * [Overview](#overview) * [Resource Files](#resource-files) * [Major Features and Updates](#major-features-and-updates) * [DRAGEN Software Mode](#dragen-software-mode) * [DRAGEN Pricing](#dragen-pricing-update) * [Map / Align](#map--align) * [Germline Structural Variant Caller](#germline-structural-variant-caller) * [UPD Caller](#upd-caller) * [Metagenomics — Epstein-Barr Virus (EBV) Detection](#metagenomics--epstein-barr-virus-ebv-detection) * [TruPath Genome](#trupath-genome) * [Somatic Small Variant Caller](#somatic-small-variant-caller) * [Somatic CNV Caller](#somatic-cnv-caller) * [Somatic Structural Variant Caller](#somatic-structural-variant-caller) * [Single-Cell RNA](#single-cell-rna) * [Bulk RNA](#bulk-rna) * [Annotation](#annotation) * [Reference Genome](#reference-genome) * [Other Updates and Bug Fixes](#other-updates-and-bug-fixes) * [Known Issues](#known-issues) * [SW Installation Procedure](#sw-installation-procedure) *** ## Overview Below is a summary of the changes included in v4.6. For full details on each feature or pipeline, please consult the latest Illumina DRAGEN Software User Guide available at . **Highlights** * **DRAGEN Software Mode** — Run DRAGEN on existing x86 servers, HPC, and public-cloud CPU instances (RHEL 8/9), with bit-exact results vs FPGA Mode. TruPath, RNA (including bulk RNA-seq), Spatial, and FastQC are not yet supported in Software Mode. * **DRAGEN Pricing** — New usage-based DRAGEN Pricing with BioInsight Platform API keys (alongside Legacy Gigabase Pricing where applicable). * **TruPath Genome** — Expanded reference support, F8 inversion genotyping, post-VC read phasing visualization, and MRJD haplotype-resolved calling including GBA. * **Germline updates** — Trio UPD caller; Epstein-Barr virus (EBV) detection and typing; germline SV precision and `` representation improvements. * **Somatic updates** — Pangenome reference support; TMB/HRD/MutSig with T/N ML (Beta); somatic CNV evidence BED, HER2 focal amplification, and arm-level events; cleaner FFPE SV calls. * **Multiomics** — Bulk RNA fusion FP reduction, PTD/ITD, systematic noise filtering, and richer fusion metrics; single-cell RNA splice junctions, guide calling, metrics, and STAR trimming improvements. * **Annotation** — DRAGEN Annotations (Annotator) v4.0: parallel workers, simplified CLI, VCF field passthrough, and UPD/methylation/JSON support. * **Map / Align** — Optional optical duplicate marking and MD tags persisted in CRAM output. Please review the section on [Known Issues](#known-issues) and limitations of the release. *** ## Resource Files DRAGEN v4.6 uses the **same resource files as DRAGEN v4.5**. No new or updated resource-file packages are required for v4.6 beyond those already published for v4.5. All resource files are available for download at the Illumina DRAGEN Product Files support site: Hash tables and other resources built for DRAGEN v4.4 or older remain unsupported. Use the v4.5 / v4.6 resource set (for example Hash Tables v12 and associated collections) as documented for v4.5. *** ## Major Features and Updates ### DRAGEN Software Mode DRAGEN v4.6 introduces **Software Mode**: run DRAGEN secondary analysis on general-purpose x86 CPU servers without a DRAGEN FPGA card. Mapping/alignment, variant calling, and FASTQ/BAM compression run entirely in software; runtime scales with thread count. Deployment options now include on-premises DRAGEN servers, Illumina cloud, public-cloud FPGA instances, **and** existing x86 servers / HPC or public-cloud CPU instances. See the user guide (*DRAGEN Software Mode*) for full requirements, install steps, and tuning. #### Supported pipelines and applications Software Mode supports a broad set of research pipelines already familiar from FPGA deployments, including: | Germline | Somatic | 5-base / single-cell | | ---------------------------------- | ------------------------------------ | ---------------------- | | SNV, Indel, SV, CNV | SNV, SV, CNV | 5-base DNA methylation | | Regions of homozygosity | Tumor-only and tumor/normal | Single-cell RNA | | Repeat expansion | TMB, MSI, MRD for WGS | Single-cell ATAC | | Targeted callers, Star Allele, SMA | Liquid biopsy, gene fusion, heme WGS | | | Trio / joint calling, imputation | | | | Specialized callers (MRJD) | | | **Not yet supported in Software Mode:** TruPath, RNA (including bulk RNA-seq), Spatial, and FastQC. Run those analyses on FPGA Mode. Application examples available with Software Mode today include WGS, WES, rare disease / family analysis / PGx, solid tumor, liquid biopsy, MRD, heme, single-cell RNA, and 5-base methylation. #### Where Software Mode can run Software Mode is intended for: * On-premises x86 HPC or commodity Linux servers * Public-cloud CPU instances (any cloud provider / geography supported by your deployment) * Nodes with outbound HTTPS to `license.dragen.illumina.com` (required at runtime on every worker node) **Software Mode cannot run in an air-gapped or dark-site environment** — there is no offline licensing option. FPGA Mode on-premises servers, Illumina cloud, and public-cloud FPGA instances remain available as before. #### Performance and runtime * **Less than 1.5 hours** for all-callers WGS 35× on an AWS `m8a.16xlarge` instance (FASTQ → BAM and all VCFs including targeted callers for PGx, STRs, and segmental-duplication genes) * **Bit-exact** FP/FN difference vs FPGA in tested comparisons
Runtime comparison: on-premises DRAGEN Server (FPGA) vs AWS EC2 F2 (FPGA) vs AWS CPU Software Mode for 35× HG002 germline all-callers

Software Mode v FPGA runtime comparison

#### Result equivalence across compute environments DRAGEN Software Mode produces the same **bit-exact** output as FPGA Mode, deterministic across thread counts and deployments. Based on internal testing of HG002, HG003, and HG004 using the same DRAGEN v4.6 analysis configuration, matching MD5 checksums were observed for SNV, CNV, and SV VCF outputs across the tested on-premises, AWS, and OCI configurations. #### System requirements and access | Requirement | Value | | ------------------------ | --------------------------------------------------------------------------------------------------------------------------------- | | Operating system | RHEL 8.x or 9.x, and RHEL-compatible Enterprise Linux | | Architecture | x86-64 with **AVX2** (mandatory) | | Threads | 16+ required; **64 recommended** (Software Mode defaults to all available cores) | | Memory | 120 GB+ required; **256 GB recommended** (WGS) | | Local storage | Fast local NVMe or SSD preferred for inputs, outputs, temp, and reference; network-attached storage supported but not recommended | | File handles / processes | `nofile` ≥ 72000; `nproc` ≥ 65535 | | Licensing | Illumina BioInsight Platform API key (metered in BioInsight Credits) | | Network | Outbound HTTPS to `license.dragen.illumina.com` at runtime | **Obtain and install:** download the self-extracting Software Mode installer `dragen-softwaremode-...bin` from the Illumina BioInsight Platform home page, run it (extracts beside the installer by default), then confirm with `/bin/dragen --version`. **Enable Software Mode** (either is sufficient): 1. **Dedicated package** — the `dragen-softwaremode` installer runs every analysis in Software Mode (no extra flag). 2. `--sw-mode` **/** `-s` — request Software Mode for a single run when not using the dedicated package. Supply the API key via `--api-key-file` (or the documented environment-variable options). See the user guide (*DRAGEN Software Mode* and *API Key Licensing*) for setup and tuning. *** ### DRAGEN Pricing Update DRAGEN uses a usage-based licensing model across most pipelines and features. Every run is licensed. The charge uses new **DRAGEN Pricing** or **Legacy Gigabase Pricing**, depending on your DRAGEN version and where you run it. | **Pricing Model** | **Description** | | ----------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- | | DRAGEN Pricing | - Every run is charged based on the type of analysis. - Enabled with API Key Licensing. - Requires DRAGEN v4.6. | | Legacy Gigabase Pricing | - Legacy on-prem and BYOL pricing model. - Usage is consumed per gigabase from a purchased quota. - Supported on all versions including DRAGEN v4.6 | * On DRAGEN 4.6, new DRAGEN Pricing is used for BioInsight Platform Core, and for BYOL using Platform API keys. On earlier versions, every deployment uses Legacy Gigabase Pricing. * API key usage draws on the same BioInsight Credit balance as your in-platform runs. * Licenses installed on a DRAGEN Server, and legacy Cloud FPGA BYOL credentials, meter against a separate prepaid gigabase quota that is not BioInsight Credits. * **Software Mode** (v4.6) requires a BioInsight Platform API key and will use new DRAGEN Pricing. Not available for air-gapped / dark-site environments. See the user guide (*DRAGEN API Key Licensing*) for more details. *** ### Map / Align #### Optical duplicate marking DRAGEN can optionally count **optical duplicates** in addition to standard PCR duplicate marking. Optical duplicates are reported as a separate count (as a percentage of PCR duplicates), which helps distinguish chemistry or library duplicates from imaging duplicates and supports more accurate library complexity estimation. An optical duplicate meets all of the following criteria relative to the primary (winner) alignment: * The read is a PCR duplicate of the primary alignment * The read is on the same tile * The read is within the configurable X,Y distance threshold on the flow cell Enable with: * `--enable-duplicate-marking=true` (required) * `--enable-optical-duplicate-marking=true` * `--optical-duplicate-distance=` (default: **100**) Tile, X, and Y coordinates are taken from the Illumina read name (QNAME). See the user guide (*Sort / Duplicate Marking — Optical Duplicate Marking*) for algorithm detail, including bounding-box and transitive grouping behavior. *** ### Germline Structural Variant Caller #### Improved inversions and translocations (BND precision) DRAGEN v4.6 improves precision for inversion and translocation calling via breakend (BND) filtering. On HG001–HG007, noise is reduced by **72.2%** for inter-chromosomal BNDs and **94.4%** for intra-chromosomal BNDs, while retaining true positives. #### Inversion VCF representation Large inversions are summarized as a single symbolic `` VCF record to aid downstream interpretation. The four underlying `BND` records are still emitted so junction-specific information is retained, but they are marked with the `Duplicate` filter. The `` record and its companion BNDs share a common `INFO/EVENT` ID (and `INFO/EVENTTYPE=INV`) so tools can group them as one event. #### INS/DEL accuracy On the HG002 NIST T2T-Q100 benchmark (T2TQ100-v1.1-v5.0q, hg38; Truvari v5.4.0), DRAGEN v4.6 improves INS/DEL precision by about **1.2 percentage points** vs DRAGEN v4.5: | Method | Precision | Recall | F-score | | ---------- | --------- | ------ | ------- | | DRAGEN 4.6 | 0.869 | 0.742 | 0.800 | | DRAGEN 4.5 | 0.851 | 0.743 | 0.793 |
Reduced noise in inter/intra-chromosomal BNDs

Reduced noise in inter/intra-chromosomal BNDs

#### Additional germline SV updates * Multi-sample SV metrics CSV now reports correct per-sample counts (previously identical across samples). * Fixed a chrM `Low_Support` false-negative case observed on a TruPath dataset. * Viral integration calls now distribute VAF appropriately. * MEI rescue path uses MEI matching status (shared with somatic SV). *** ### UPD Caller DRAGEN v4.6 adds a trio-based **uniparental disomy (UPD)** caller for genome-wide detection of UPD from germline sequencing. UPD events (both chromosome copies inherited from one parent) are often missed by standard variant callers and are relevant to rare disease and imprinting-related disorders.
What is UPD?

What is UPD?

#### Trio-based UPD detection * Designed for **trio WGS and WES** * Detects UPD from trio genotype patterns, including **whole-chromosome** UPD * Distinguishes **isodisomy (ISO)** from **heterodisomy (HET)** using LOH * Runs downstream of single-sample germline calling (three gVCFs/VCFs) or on a joint-genotyper multisample VCF Enable with `--enable-upd-caller=true` and a pedigree file (`--pedigree-file`) for exactly one three-member family. Recommended inputs include `--upd-loh-cyto` (proband cyto LOH VCF) and `--upd-proband-cnv` (to mask aneuploid regions). See the user guide (*UPD Caller*) for full options. #### UPD output files and VCF fields Primary output: `.upd.vcf.gz` (one `` record per event). Additional outputs: `.upd.table.tsv` and `.upd.probe.bed.gz`. Key FORMAT fields include: | Field | Meaning | | --------- | ---------------------------------- | | `UPDTYPE` | `ISO` / `HET` / `COMPLEX` | | `UPDPAR` | Parental origin (`MAT` / `PAT`) | | `UPDSIG` | Statistical significance (p-value) | | `UPDVAR` | Supporting variant sites | | `UPDBPI` | BPI sites on the chromosome | Segment span is reported via `POS`–`END` and `SVLEN`. Example (maternal heterodisomy of chromosome 20): ``` chr20 101498 UPD:chr20:101498:64225348 N . PASS END=64225348;SVLEN=64123850 GT:UPDVAR:UPDBPI:UPDTYPE:UPDPAR:UPDSIG 0/1:8350:31:HET:MAT:0 ``` *** ### Metagenomics — Epstein-Barr Virus (EBV) Detection DRAGEN v4.6 adds a research add-on for **Epstein-Barr virus (EBV)** detection and Type 1 vs Type 2 typing on germline WGS, WES, and panel workflows. The EBV database is packaged with DRAGEN (no separate download). Viral genome content can be sequenced alongside the human genome without targeted amplification. #### EBV detection and typing pipeline * EBV is associated with cancers and autoimmune diseases * Seamless add-on to existing germline workflows (no database download) * Infects >90% of the global population (highly ubiquitous) * Type matters: Type 1 vs Type 2 differ in biology and global distributions * Pipeline * DRAGEN map/align * K-mer classification of non-human reads (unmapped reads; reads mapping to decoy `chrEBV` are included when present in the reference) * EBV detection from genome coverage signals * Subtype resolution (Type 1 or Type 2) via EBNA2/EBNA3 genotyping Enable with `--enable-human-microbe-detection=true`. Not compatible with read collapsing / UMI. See the user guide (*Epstein-Barr Virus Detection*) for full options. Primary output: `.microbe_detections.ebv.tsv` (RPKM, read count classified to the virus, and percent of the most likely reference covered at ≥1×). Simplified example below (HG02790, 1000 Genomes blood sample, EBV Type 1): | Microbe | Detected | RPKM | Read Count | % viral reference ≥1× | | ------------------------------- | -------- | ----- | ---------- | --------------------- | | Epstein-Barr virus (EBV) | Y | 0.036 | 1036 | 60.2 | | Epstein-Barr virus (EBV) Type 1 | Y | 0.053 | 193 | 60.5 | | Epstein-Barr virus (EBV) Type 2 | N | — | — | — | #### Validation against orthogonal assays On 128 samples compared to an orthogonal assay/informatics approach (not Illumina-affiliated; does not report EBV type): * High correlation in EBV supporting reads vs the orthogonal approach * All 24 orthogonal EBV+ samples were typed by DRAGEN (including Type 2) * DRAGEN was more sensitive than orthogonal EBV− calls (presence confirmed by additional bioinformatics); type resolution improves with increasing viral read count
High correlation in EBV supporting reads between DRAGEN and an orthogonal approach while also reporting EBV type

EBV supporting reads correlation

*** ### TruPath Genome DRAGEN v4.6 continues TruPath Genome improvements spanning expanded reference support, F8 inversion genotyping, post-VC read phasing visualization, and Multi-Regional Joint Detection (MRJD) haplotype-resolved calling (including GBA). #### Detection of F8 inversions TruPath colocation information is used so common **F8** intron 1 and intron 22 inversions (\~half of severe hemophilia A cases) are considered during SV calling. Results appear in the SV VCF as one `` record plus four companion `BND` records. | Type | chrX ploidy | Sensitivity | Specificity | | --------- | ----------- | ------------ | ------------ | | Intron 22 | Haploid | 100% (16/16) | 100% (23/23) | | Intron 1 | Haploid | 100% (1/1) | 100% (38/38) | | Intron 22 | Diploid | 100% (2/2) | 100% (28/28) | | Intron 1 | Diploid | No data | 100% (30/30) |
Sample with F8 inversion

Sample with F8 inversion

Negative control

Negative control

#### Post-VC read phasing visualization DRAGEN v4.5 TruPath BAMs showed **pre-VC** read phasing—the phasing used as input to variant calling. At some sites that left fewer reads tagged as phased in the pile-up, which can make visual inspection misleading compared with typical third-party phasing views. DRAGEN v4.6 improves BAM visualization by showing **post-VC** read phasing (often called haplotagging): after variants are called and phased, reads are reassigned to haplotypes so pile-up haplotype tags match the phased variant calls. The result is more complete pile-ups with more phased reads, comparable to other phasing pipelines that assign reads to haplotypes after phasing variants. What changed vs v4.5: * BAM shows post-VC haplotype tags that agree with phased variant genotypes * Read pile-ups are more complete for visual review of call support * Enabled by default for TruPath (`--enable-proximity=true`) What did **not** change: * Variant calling and variant phasing logic are unchanged—those already used the full phasing information * Long-range TruPath / personalization phasing upstream of VC remains as before BAM tags (post-VC defaults): * `HP` — haplotype tag (`1` / `2`) for reads with sufficient phasing confidence * `pp` — Phred-scaled phasing probability (log odds) * Optional pre-VC scores as `HZ` / `pz` via `--vc-include-raw-read-phase-scores` (mirrors the post-VC `HP` / `pp` tags)
Post-VC read phasing

Post-VC read phasing

See the user guide (*Phasing* under *DRAGEN Germline Pipeline for Illumina TruPath Genome*) for options and output details. #### Haplotype-resolved variant calling in GBA (MRJD) TruPath extends MRJD with copy-number–aware, haplotype-resolved small variant calling across 16 disease-associated paralog regions, now including **GBA**. Outputs include haplotype-resolved VCF, phased BAM, gene-specific copy number, and haplotype visualization files. * Supported gene/region: PMS2; SMN1/SMN2; NCF1; RCCX (CYP21A2, TNXB); GBA; STRC; CYP2D6; CYP11B1/CYP11B2; CFHR1–4; USP18. * TruPath MRJD enables finding recombination *de novo* in *GBA*
GBAP1-GBA hybrid in haplotype 2

GBAP1-GBA hybrid in haplotype 2

#### MRJD accuracy and robustness updates MRJD model robustness and accuracy are improved across paralogous genes, delivering highly accurate phased variant calling. Haplotype concordance vs on-market long-read results on 14 cell lines with diverse ancestry (HMW and standard DNA): | Paralogous gene | Disease research relevance | HMW DNA mean concordance | Standard DNA mean concordance | | --------------- | --------------------------------------- | ------------------------ | ----------------------------- | | PMS2 | Lynch Syndrome | 0.988 | 0.987 | | SMN1–SMN2 | Spinal Muscular Atrophy | 0.933 | 0.927 | | NCF1 | Chronic Granulomatous Disease | 0.969 | 0.970 | | CYP21A2 | Congenital Adrenal Hyperplasia | 0.999 | 1.000 | | TNXB | Ehlers-Danlos Syndrome | 0.999 | 1.000 | | GBA | Gaucher Disease | 0.985 | 0.984 | | STRC | Recessive Nonsyndromic Hearing Loss | 0.980 | 0.978 | | CYP2D6 | Pharmacogenetics | 0.977 | 0.978 | | CYP11B1–CYP11B2 | Glucocorticoid-remediable Aldosteronism | 0.998 | 0.998 | Long-read baseline not available for CFHR1–4 and USP18. #### Expanded reference genome support TruPath support varies by reference genome: **hg38:** Full variant class calling capabilities * Mapping, small variants, SV, CNV, STR, MRJD, colocation, visualization * Baseline for all TruPath capability claims **hg19 & CHM13:** Selected workflows only * Supported for selected TruPath workflows, not feature parity with hg38 * STR and MRJD analysis are not available * Visualization unavailable (see capability matrix) **Diploid non-human:** Linear diploid refs only * Any diploid non-human linear reference from the DRAGEN hash table builder * Polyploid genomes are not supported * Colocation mapping is not supported | Capability | hg38 | hg19 | CHM13 | Custom human | Diploid non-human† | | ----------------------------- | ---- | ---- | ----- | ------------ | ------------------ | | Mapping (proximity) | Yes | Yes | Yes | Yes | Yes | | Small variant calling | Yes | Yes | Yes | Yes | Yes | | SV calling | Yes | Yes | Yes | Yes | Yes | | CNV calling | Yes | Yes | Yes | Yes | Yes | | STR calling | Yes | No | No | No | No | | MRJD (paralogs) | Yes | No | No | No | No | | Colocation mapping | Yes | Yes | Yes | Yes | No | | Platform Core / Visualization | Yes | No | No | No | No | | DRAGEN read phasing | Yes | Yes | Yes | Yes | No | | Mosaic calling | Yes | Yes | Yes | Yes | No | | IRR recovery (STR) | Yes | Yes | No | Yes‡ | No | | `--read-phasing-gene-list` | Yes | Yes | No | Yes‡ | No | Custom human linear refs via DRAGEN hash table builder (not pangenomes); baseline mirrors hg19. ‡ IRR / gene-list may follow CHM13 rules depending on build. † Diploid non-human linear only — polyploid excluded. Colocation = No in product 4.6. Additional MRJD / TruPath updates: * Improved handling of mismapped reads in genotyping (BED update) * Bounding-box inclusivity and overlapping-event abort behavior fixes * Indel alignment and invalid fragment-pair handling improvements * Non-tandem checksum consistency when flank information is absent * Proximity model excludes `chrEBV` from model fitting *** ### Somatic Small Variant Caller #### Pangenome reference support for somatic workflows * Somatic pipelines now accept **linear or pangenome** human references with no change to defaults or enforcement. Existing linear-based somatic workflows are unchanged. * Using the same pangenome reference for tumor and normal enables coordinate-consistent tumor/normal analysis when the normal was processed through a germline pangenome workflow. * Pangenome references primarily benefit germline analysis workflows without degrading accuracy on DRAGEN somatic applications * See [Reference Genome](#reference-genome) for recommended-reference guidance. #### TMB, HRD, and MutSig with Somatic T/N ML (Beta) * DRAGEN v4.5 introduced **Machine learning (ML) somatic variant calling for Tumor/Normal (Beta).** * DRAGEN v4.6 extends support for TMB, HRD, and MutSig biomarkers when Somatic tumor/normal machine-learning (T/N ML) filtering is enabled. * See the user guide (*Somatic ML for Small VC (Beta)*) for details. #### Additional somatic SNV updates * Tumor/normal calling no longer fails on IUPAC ambiguous reference bases * T/N ML filtering skips only germline events (avoids over-filtering somatic candidates) *** ### Somatic CNV Caller #### Input evidence BED for somatic CNV Optional `--cnv-evidence-bed=` supplies expected total copy number per region (fourth BED column) from orthogonal assays such as FISH, cytogenetics, or prior assessments. * A standard BED file indicating expected total CN, for genomic regions already assessed with orthogonal technologies, can be provided as input to DRAGEN Somatic CNV and constrain the model selection only to compatible candidates * Improves model selection confidence * Incorporates orthogonal laboratory evidence * Supports more accurate CNV interpretation * Best for labs with existing orthogonal CN data that want tighter CNV calls without a separate interpretation step Example: BED file with expected CN in fourth column, formatted similarly to: ``` chr8 91954967 92103385 6 chr21 34787801 36004667 8 chr16 67028984 67101058 4 ``` #### Focal gene amplification detection DRAGEN v4.6 extends copy number analysis from genome-wide events to focal gene amplification assessment directly from sequencing data. * Focal gene amplification assessment (including HER2/ERBB2) is available from sequencing data alone, alongside standard CNV analysis. * HER2 amplification classification evaluated on 623 WES tumor-only breast cancer research samples vs CAP IHC/FISH: Sensitivity **95%** (59/62), Specificity **99%** (538/544), No-call rate **3%** (17/623) * Outputs include HER2 copy estimate (`HER2CN`) and FISH-style ratio vs a control region (`HER2Ratio` / `SMRATIO`). See the user guide (*Focal gene copy number assessment*). * Best for research workflows evaluating focal amplification events as part of broader genomic characterization studies
Example HER2 amplification

Example HER2 amplification

#### Arm-level and whole-chromosome CNV detection DRAGEN v4.6 identifies arm-level and whole-chromosome copy number events directly without needing any post-processing or custom scripts. * Detect large chromosomal gains and losses directly within DRAGEN CNV analysis * Enable via `--cnv-enable-arm-level-events=true` (**default false**). * DRAGEN reports a default WHO-relevant arm set (`1p`, `1q`, `3p`, `3q`, `6p`, `6q`, `7p`, `7q`, `10p`, `10q`, `14q`, `18p`, `18q`, `19q`, `22q`), including events such as 1p/19q co-deletion. * Add custom arms/regions with `--cnv-segmentation-bed`. * Results are generated as part of standard CNV outputs * Reduces the need to manually aggregate segment-level CNVs into chromosome-arm events * Best for neuro-oncology and other research applications evaluating large-scale chromosomal CNAs #### Additional somatic CNV updates * FORMAT scoring for subclonal CNV events. For HET/MOSAIC CNVs, adds FORMAT HLR (somatic) / MLR (germline) — log10 LLR favoring heterogeneity — plus CQ/NMQ (clonal quality), and fixes writing LR. * BAF+depth union segmentation is now enabled on the WES somatic path * Depth+BAF union segmentation defaulted on for somatic WGS to reduce missed CNLOH * TSO500 solid/liquid CNV counts preserve original target BED names (needed for gene-scaled MAD) Shared CNV improvements that also apply to germline are listed under [Other Updates and Bug Fixes — CNV](#cnv). See the user guide (*Somatic CNV Calling*) for more information. *** ### Somatic Structural Variant Caller #### Cleaner structural variant calls from FFPE samples DRAGEN v4.6 reduces somatic SV false positives for FFPE and tumor-only workflows (more restrictive incomplete-insertion matching; relaxed systematic-noise matching in noisy regions for tumor-only WGS) while improving consistency across FFPE samples. On 8 FFPE clinical WGS T/N samples (\~90×, 2 replicates): * **61%** fewer replicate-unique PASS calls (T/N) * **64%** fewer replicate-unique PASS calls (tumor-only) * **53%** reduction in tumor-only noise vs matched T/N
Reduction of FPs on somatic FFPE data

Reduction of FPs on somatic FFPE data

#### Additional somatic SV updates * SV hotspot support is extended for deletions (increased sensitivity in configured hotspot regions). * Somatic SV calling uses less memory for supported use cases. * The somatic joint CNV/SV rearrangement output is named `*.rearr.vcf.gz` (original `*.cnv.vcf.gz` and `*.sv.vcf.gz` remain available). * `--sv-report-small-dup-as-ins` now defaults to **false** (small duplications are not converted to insertions by default). See the user guide (*Structural Variant Calling*) for more information. *** ### Single-Cell RNA #### Splice junction counts in count matrices Optional splice-junction counting (`--scrna-count-splice-junctions=true`; disabled by default) adds annotated GTF junctions to the cell×gene matrix, or emits separate `*.scRNA.sj_matrix.*` files with `--scrna-output-split-matrices=true`. * Best for single-cell RNA studies investigating transcript variation, RNA processing, or cell-specific splicing patterns * Investigate RNA processing and potential alternative-splicing patterns across cells * Choose between an integrated count matrix and dedicated splice-junction outputs * Add junction-level information without creating a separate analysis workflow #### Standalone cell filtering and guide-calling rerun Re-run cell filtering and guide calling from an existing raw count matrix without repeating mapping, alignment, deduplication, or counting. * Best for single-cell studies that require comparison or optimization of cell-filtering and guide-calling settings * Compare different cell-filtering strategies using the same upstream results * Adjust cell-count thresholds and guide-calling parameters more efficiently * Avoid unnecessary reprocessing when only filtering settings need to change Usage: * Enable with `--single-cell-enable-cell-filter=true` * Required inputs: `--single-cell-filter-input-directory=` , `--single-cell-filter-input-prefix=` * Reads the input files that are present: `.scRNA.barcodes.tsv.gz` , `.scRNA.features.tsv.gz` , `.scRNA.matrix.mtx.gz` * Configurable filtering options: `--single-cell-threshold=` , `--single-cell-number-cells=` * Optional guide calling: `--scrna-guide-calling-mode=gmm` #### Granular read-group and feature-type metrics Expanded feature reporting and dedicated read-group and feature-type metrics provide more granular insight into single-cell experiment performance * Best for single-cell experiments combining gene expression with antibody, guide, or other feature-barcode measurements * Review quality metrics at the individual read-group level * Compare performance across gene-expression and configured feature types * Gain additional context for evaluating feature detection and assay quality When feature counting is enabled (`--scrna-feature-barcode-reference=`), DRAGEN expands `*.scRNA_metrics.csv` and writes new outputs `*.scRNA.read_group_metrics.csv` and `*.scRNA.feature_type_metrics.csv`. | Metrics reported | v4.5 | v4.6 | | ----------------------------- | ---- | ---- | | High-level QA metrics | Yes | Yes | | Expanded feature metrics | No | Yes | | Per read-group metrics file | No | Yes | | Per feature-type metrics file | No | Yes | #### Guide assignment context and paired-guide reporting More guide-assignment context, an additional threshold-based output, and dedicated reporting for user-defined combinations of two or more guides * Best for CROP-seq and other single-cell perturbation studies evaluating individual guides or defined guide combinations * Compare complementary approaches to individual-guide assignment * Use posterior probability to add context to guide calls * Identify cells containing complete, user-defined guide combinations Guide calling adds: * GMM posterior-probability outputs, * a simple global count-threshold file (default threshold 10; `--scrna-guide-calling-global-count-threshold`), and * optional paired-guide reporting (`--scrna-enable-paired-guide-calling=true` with `guide_pair_id` in the feature barcode reference; all guides in a pair must pass). Option renamed to `--scrna-guide-calling-mode` (legacy `--scrna-crispr-guide-calling-mode`). Outputs include `positive_guide_assignments.csv`, `simple_threshold_guide_assignments.csv`, and paired equivalents. #### STAR-based scRNA adapter and homopolymer trimming Adapter and homopolymer trimming is now automatically applied in STAR-based scRNA software-mode workflows, bringing preprocessing closer to DRAGEN hardware-mode PIPseq behavior * Best for PIPseq and other supported scRNA workflows using the STAR mapper in DRAGEN software mode * Reduces the need to configure equivalent trimming separately * Supports more consistent preprocessing between applicable software- and hardware-mode workflows * Automatically handles the listed TSO, PCR, poly-A, and poly-G sequences With `--single-cell-enable-star-mapper=true`, TSO/PCR adapter and poly-A/poly-G homopolymer trimming is applied automatically (closer to FPGA/HW PIPseq trimming). | | | | | ------------------------ | -------------------------------------------------- | -------------------------------------------------- | | **Trimming** | **Sequence** | **STAR Options Added** | | **TSO and PCR adapters** | **TSO:** AGAGTGAATGGG **Smart\_PCR:** TCAACGCAGAGT | --trim-adapter-r1-5prime=star\_adapters.fasta | | **Homopolymers** | **polyA:** AAAAAAAAAAAA **polyG:** GGGGGGGGGGGG | --trim-adapter-read1=star\_homopolymer\_trim.fasta | See the user guide (*Illumina scRNA Prep*) for more information. *** ### Bulk RNA #### Fusion false-positive reduction DRAGEN v4.6 reduced fusion false-positive calls by about **40%** vs v4.5 while retaining the same true-positive count, in the evaluated dataset (179 samples; 531 expected fusions; 400 detectable by at least one caller). Evaluation was without a systematic noise file — further FP reduction is possible with the noise workflow below.
Accuracy across versions and vs 3rd parties

Accuracy across versions and vs 3rd parties

* Best for whole-transcriptome RNA-seq studies that need focused fusion detection while retaining configurable discovery options * Produces a more focused set of fusion candidates for review * Retains study-specific flexibility through configurable gene filtering * Reduces unwanted calls primarily in whole-transcriptome sequencing analysis #### Targeted fusion gene list (WTS and enrichment panels) Primary driver of the WTS FP reduction. * Any fusion call requires at least one of the genes in the pair to be in the list. * WTS uses a default cancer-relevant list; panels use the panel gene list/BED. Override with a custom list or disable via `--rna-target-genes=none`. #### Fusion reporting by gene biotype Configure which gene biotypes are considered for fusion reporting, with protein-coding and lincRNA selected by default. * Both fusion partners must match selected biotypes. * Default: `protein_coding,lincRNA`. Use `--rna-gf-include-biotype=all` for unrestricted reporting (also supports miRNA, shortRNA, rRNA, immune, pseudogene keywords — see user guide). #### Additional fusion output metrics (FFPM, transcript, exon, expression) Fusion VCF outputs include normalized abundance, transcript, exon, expression, and read-count information for more informed downstream analysis. * Best for bulk RNA-seq workflows requiring detailed fusion review, downstream annotation, or comparison across samples * Compare fusion-supporting abundance across samples or runs * Examine transcript and exon context around fusion breakpoints * Connect fusion calls with expression information for both gene partners New output fields: * FFPM: non-duplicated fusion-supporting fragments per million reads * Transcript IDs: Transcript information for each gene partner * Exon numbers: Exon context for each fusion breakpoint * With `--enable-rna-quantification=true`; TPM: expression value for each gene partner; NumReads: Read count associated with each gene partner. * *Note*: Intronic events may report neighboring exon intervals (for example `3-4`). These fields support annotation but do not directly determine in-frame status. #### PTD and ITD self-fusion detection Configurable PTD/ITD self-fusions when a fusion systematic noise file is used * Requires fusion systematic noise file to mitigate high FPR. Auto-enabled with the noise file unless `--rna-gf-ptd-genes=none`. * Default genes: BCOR, FGFR1, FGFR2, FLT3, KIT, KMT2A, PDGFRA, UBTF. * Marked with `ITD_PTD` in the fusion VCF. Validated on KMT2A PTD (AML) and FGFR1 kinase-domain duplication examples. #### Systematic noise filtering for fusions and splice variants DRAGEN v4.6 uses recurrent events detected across normal samples to annotate and filter likely technical noise from fusion and splice-variant results * Best for RNA-seq studies with access to representative normal samples generated using comparable library preparation and analysis settings * Produces a more focused set of RNA events for downstream review * Accounts for recurring noise associated with an assay or workflow * Supports both gene-fusion and splice-variant analysis Three-step assay-matched normals workflow: | Step | Gene fusion | Splice variant | | --------------- | -------------------------------------------------- | -------------------------------------------------------------- | | 1. Detect noise | `--rna-gf-detect-systematic-noise=true` | `--rna-splice-variant-detect-systematic-noise=true` | | 2. Build BEDPE | `--rna-gf-build-systematic-noise-vcfs-list=` | `--rna-splice-variant-build-systematic-noise-vcfs-list=` | | 3. Apply | `--rna-gf-systematic-noise=` | `--rna-splice-variant-systematic-noise=` | * Overlapping events are tagged `InSystematicNoiseFile`; low VAF/alt-count overlaps receive `SystematicNoise`. * Prefer ≥30 matched normals from the same library prep. Too few normals may leave noisy ITD/PTD calls — disable with `--rna-gf-ptd-genes=none` if needed. #### Splice-variant noise reduction with systematic noise With systematic noise filtering: * \~**70%** fewer reported artifacts/unwanted splice-variant calls while preserving expected event detection * **100%** positive concordance vs a third-party commercial RNA pipeline on known FFPE tumors; * Sensitivity maintained on TSO500 LOD dilutions (100%, 50%, 25%, 10%).
Overall calls decrease after systematic-noise filtering

Overall calls decrease after systematic-noise filtering

*** ### Annotation DRAGEN v4.6 bundles **DRAGEN Annotations (Annotator) v4.0**. External docs: . On-premises path: `/opt/dragen//share/nirvana/` (cloud: `/opt/edico/share/nirvana/`). #### Faster annotation with parallel workers Parallel worker model improves throughput (\~**70%** faster with 20 workers on 17 WGS VCFs). Configure with `--parallel.workers` (default: CPU count). #### Simplified Annotator CLI and configuration Single `Annotator` executable with subcommands (`setup`, `download`, `annotate`, `list`, `version-validate`) and configuration-driven RunConfig JSON (replaces the prior multi-utility Data Manager / Downloader pattern). #### VCF INFO and FORMAT field passthrough All input VCF INFO and FORMAT fields are passed through to annotated output for consistent annotation across VCF representations. #### UPD, methylation, and JSON output support Expanded support for UPD annotation (ISCN parental notations), optional methylation annotation, and full JSON output (`--output.format json`) in addition to annotated VCF. *** ### Reference Genome #### Recommended reference by analysis pipeline v4.6 updates the recommended-reference table (including expanded pangenome support for somatic and related pipelines). Somatic accuracy characterization remains against linear references; pangenome is fully supported for somatic when coordinate consistency is desired. Hash tables remain the v4.5 / v4.6 resource set (HTv12). **Table: v4.6 Reference Support and Recommended Use for Human Data** | Pipeline | hg19 | hs37d5 | hg38 | chm13 | Recommended (human) | | -------------------------- | ------- | ------- | ---- | ------- | ------------------- | | **Germline** | Yes | Yes | Yes | *Note1* | Pangenome | | **Somatic** | Yes | Yes | Yes | *Note2* | Linear or Pangenome | | **RNA** | Yes | Yes | Yes | *Note1* | Linear | | **Methyl 5-base Germline** | Yes | Yes | Yes | No | Pangenome | | **Methyl 5-base Somatic** | Yes | Yes | Yes | No | Linear or Pangenome | | **Methyl TruSeq** | Yes | Yes | Yes | No | Linear | | **scRNA** | Yes | Yes | Yes | *Note1* | Linear | | **TruPath** | *Note3* | *Note3* | Yes | No | Pangenome | | **Annotation** | Yes | Yes | Yes | No | Pangenome | Note1 — Component execution supported; accuracy not fully established for all callers (see user guide). Note2 — Component execution supported; accuracy not fully established for all somatic assays on chm13. Note3 — Experimental use only on hg19/hs37d5. To suppress the linear-reference validation error when a pangenome reference is recommended, set `--validate-pangenome-reference=false`. *** ## Other Updates and Bug Fixes Minor updates and customer-facing bug fixes included in DRAGEN v4.6 (organized by component). ### Alignment / Map-Align * Fixed 0D liftover CIGAR handling in graph/pangenome runs that could crash the SV caller * When MD tag generation is enabled (`--generate-md-tags=true`), DRAGEN now persists MD tags in CRAM output as well as BAM. ### Amplicon * Improved amplicon indel calling at target boundaries, including fixes for missed CALR 52 bp deletion on MPN/Myeloid panels and Pillar false negatives at target edges. * Count ref-matching soft-clip bases for column-wise depth (DP) ### BCL Convert * BCL Autodetect updates for supported instruments and sample sheets: per-lane index correction (including index1/index2 swap), Sample\_Project consistency checks, and `IndexDetectResult` in `AutoDetect_Details.json`. * High Yield Mode for BCL conversion: set sample-sheet `FilterMode=HighYield` (optionally per lane), with optional `IncludeReadQualityInHeaders` and IntensitiesRQI run-folder support for read-quality headers. * See the user guide (*BCL Autodetect and Correct*) for details. ### CNV * Improved WGS sex estimation for partial chrY deletions: samples with residual Y signal are no longer called X0 based on low median Y coverage alone (WES logic unchanged). * Sex genotyping and ploidy estimation are consolidated so CNV prefers PloidyEstimator results when available and records sample ploidy in target/GC counts headers. * Germline ASCN now applies the same high copy-number (`highCN`) and likelihood-ratio (`cnvLikelihoodRatio`) filters as depth-only germline CNV, with FORMAT `LR` when enabled (`--cnv-filter-high-cn`, `--cnv-filter-likelihood-ratio`). * Updated `cnv-segment-max-gap` default for more reliable segmentation * Preserve original TSO500 target BED names in CNV counts * Germline CNV calls can be rescued by matching to SV breakends for complex events (for example DEL-INV), including POS adjustment and `SVCLAIM=D` on matched non-REF CNVs. * Fixed germline WGS ASCN whole-genome trisomy / triploidy detection when no segments have \~50% MAF for model fitting (can now call genome-wide CN3). * Shared CNV improvements (germline and somatic): omit CIPOS/CIEND for bed-segmentation CNV records; more consistent CNV VCF QUAL across CPU models; soft failure (warning + empty outputs) instead of a crash on empty GC-bias input. * Avoid crash on chrY when a male sample has no chrY intervals * Default depth+BAF union segmentation for somatic WGS (reduce missed CNLOH) * QUAL scoring for subclonal events * Enable BAF+depth union segmentation on WES somatic path ### 5-Base / Methylation * Improved Methyl-Seq / 5-base active-region filtering reduces runtime while preserving accuracy (conversion-like bases treated as matches when defining active regions). * **Methylation Base Masking (Beta):** exclude bases at read ends from methylation reporting only (XM, CX report, M-bias, and related metrics) without trimming reads for map/align or variant calling. Set with `--methylation-reporting-exclude-read-bases` as `5'R1,3'R1,5'R2,3'R2` (e.g. `0,0,9,0`; default `0,0,0,0`). * Fixed fragmentomics end-motif calculation for Illumina 5-base conversion (`--methylation-conversion=illumina`): conversion artifacts are remapped before motif counting so end motifs are not biased toward TGTT/CATT. ### Germline Small Variant Caller * No major new germline small-variant caller features are introduced in v4.6 beyond related pipeline updates described under [TruPath Genome](#trupath-genome), [UPD Caller](#upd-caller), and [Reference Genome](#reference-genome). ### HLA * Fixed an HLA analysis crash from an uninitialized duplicate watchdog ID. ### iGG / Population genotyping * Pedigree caller FORMAT metrics **DN** and **DQ** are ingested into iGG multi-sample VCF outputs when present on input multi-sample gVCFs. * Cohort ML scores are written to full msVCF INFO as **MLQUAL** / **MLFILTER** (per-allele; comma-joined at multi-allelic sites) when cohort ML is enabled. * Added `--gg-universal-sharding` for equal-length sharding on non-human genomes (default false; see user guide for interaction with `--gg-v1-sharding` and contig include options). * Filtering option for non-ACGT alleles in germline gVCF inputs * Additional iGG fixes: genotype phase flip between gVCF and multi-allelic msVCF; relative paths accepted for iterative gVCF inputs; prevent step-3 assertion on GATK gVCFs with AC=0 / spanning-deletion alleles. * Fixed cohort ML GQ binning at multi-allelic sites (improves FILTER/QUAL decisions at those sites). ### MRJD / TruPath * Non-tandem checksum consistency when flank information is absent * MRJD bounding box treated as inclusive * Abort correctly when too many overlapping MRJD events occur * Indel alignment improvements * Treat fragment pairs with one disqualified mate as invalid ### Oncovirus Detection * Fixed oncovirus detections TSV trailing-tab formatting and a multi-thread race that could corrupt `.read_classifications.tsv` when `--oncovirus-detection-enable-read-output` is used. * Best-match reference read counts in `*.oncovirus_detections.tsv` (`best_match_ref_read_count`) now include reads that tie for best match, not only unique best matches. ### Paralog / Ploidy / Proximity / Targeted * Most targeted callers are disabled by default to support Foundation pricing (except Rh, which remains on). Callers that no longer auto-enable with VC include SMN, GBA, CYP21A2, CYP2D6, and HBA — enable explicitly with `--enable-targeted=` or `--enable-targeted=true`. See the user guide for defaults and supported targets. * The star allele caller detects whole-gene deletion of **UGT2B17** (reported as `*2` when a large overlapping DEL spans the gene). * Force per-record SV genotype consistency with joint genotype (paralog caller) * Fix `--pe-coverage-factors` indexing on hg19; default WES Y coverage factor to 0.8 in Ploidy Estimator * Proximity mapping output consistency between local and AWS runs * Do not report an empty set of small-variant calls from targeted callers ### Platform * DRAGEN v4.6 is ready for upcoming Oracle Linux 9 and DRAGEN Server v5 platforms. ### QC Metrics * Q-score histogram metrics / telemetry are computed from input FASTQ before UMI collapsing (UMI-only FASTQs are ignored for the histogram). ### RNA (bulk) * Edge-case fusion sensitivity and LOD / low-support filtering improvements * Fix latent BufferPool use-after-destroy in RNA splice variant caller * Reduce stderr warnings when splice fusion genes are absent from the GTF * Fix incorrect intergenic fragments percentage in metrics * Fix preliminary vs final fusion-pair file consistency * Intermittent RNA VC crash fix (SKIP cigar path) * Splice-variant false-negative fixes vs orthogonal alt-splice expectations * Fusion calls discordant with orthogonal pipelines; target-list update restoring TP counts * Add IKZF1 exon 3–7 deletion to splice knowns * CIPOS reported for fusions without assembly * Fix crash in RNA splice "Fatal error: Invalid argument" ### Single-Cell RNA * Split large bins when partitioning by cell barcode * Fix barcode phasing and whitelist interaction * CRISPR/feature mapping for nonstandard positioning * Fix off-by-one h5ad count matrix issue * Rename `--scrna-split-counts-by-barcode` to `--single-cell-group-by-cell-barcode` * Fix ambiguous normal cells in tumor-normal demultiplexing * Fix scRNA tumor-normal cell demultiplexing accuracy failure ### Other * Spatial analysis handles ambiguous / trans-spliced transcript strand annotations for non-eukaryotic and agrigenomics references. *** ## Known Issues The following known issues exist in DRAGEN v4.6. Where applicable, a workaround is provided. | Component | Description | Workaround | | --------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **SV** | DRAGEN SV may rarely hang and abort with a watchdog timeout after locus merging completes. This is a very rare race condition. | No workaround. Re-run will pass. A fix is planned for a future release. | | **SV** | When merging two different SV types at the same location, the SV caller may occasionally fail to classify `SVTYPE` correctly. This is uncommon. | Bypass the affected path with `--sv-enable-pre-assembly-merging=false`. A fix is planned for a future release. | | **BCL Convert** | `AutoDetect_Details.json` may intermittently record an empty `DetectedOverrideCycles` field for lane 1. This is a logging/details-file issue and does not affect demultiplexing of reads. | No workaround. A fix is planned for a future release. | | **Gene Fusion** | Gene fusion VCF entries for splice variants may have an empty `MATEID` field. | No workaround. A fix is planned for a future release. | | **QC Metrics** | In tumor/normal runs, the Normal mapping metric *Input bases divided by reference genome size* may use bases from both samples instead of the normal sample only, producing an inflated value in `mapping_metrics.csv`. | Ignore this Normal coverage-depth proxy metric, or compute depth from Normal total bases ÷ reference size manually. A fix is planned for a future release. | | **Somatic SNV** | On pangenome graph references, mitochondrial (mito) variant calling and allele-fraction reporting may be affected by reduced coverage when reads are recruited to graph population-alt contigs instead of the mito contig. | Prefer a linear reference when mito accuracy is critical, or treat graph-reference mito results with caution. A fix is planned for a future release. | | **Map / Align** | Very rare hardware run-to-run variation has been observed, limited to the `hl:Z` BAM tag; other mapping fields are unaffected. | No workaround. Documented as a rare possibility; a hardware-side fix is not available for v4.6. | | **CNV** | Whole-chromosome cytogenomic CNV genotype (`GT`) fields may differ between AWS and on-premises runs (for example `./1` vs `1/1`) while the DUP call and other fields remain consistent. This is caused by floating-point non-determinism in the CNV model-fitting stage surfacing on a borderline/low-confidence whole-arm segment. Affected records are typically filtered (not PASS). | No workaround. Impact on PASS calls is expected to be low. A fix is planned for a future release. | | **Metagenomics** | If a worker task raises an exception, DRAGEN may hang instead of exiting with a clear error. This should be infrequent. | No workaround. A fix is planned for a future release. | | **Software Mode, 5-base Somatic** | On 5-base somatic datasets, Software Mode and FPGA Mode may show rare non-bit-exact differences in fields in SNV VCF output. Differences are typically limited to one or two VCF records. | No workaround. Treat Software Mode and FPGA Mode as functionally equivalent for 5-base somatic variant calls. | | **5-Base / Methylation** | In somatic 5-base workflows, methylation reporting in VCF/gVCF at SNV (variant) positions is disabled because prior reporting could be inaccurate. | No workaround for per-variant methylation in somatic VCF/gVCF in v4.6. Re-enablement is planned for a future release. | | **5-Base / Methylation** | Methylation reporting (`M5mC`) at somatic multiallelic sites can be incorrect. | No workaround. A fix is planned for a future release. | | **Germline SNV** | Intermittent run-to-run differences may appear in hard-filtered VCF checksums (and related personalization outputs). Differing fields are typically small numeric or allele-order differences and do not change PASS vs filtered status. This behavior was also present in v4.5. | No workaround. | | **Germline SNV** | Some variants may have a lower QUAL score than expected based on read support and mapping quality. These variants are still called correctly; the QUAL discrepancy is related to ML training data and does not affect overall accuracy. This was also present in v4.5. | For information only. No workaround required. | | **Somatic SNV / UMI** | Tumor+UMI/Normal workflows from BAM or CRAM input with `--tumor-normal-has-umi=tumor` may crash. FASTQ-input workflows are unaffected and are the recommended input mode. This was also present in prior releases. | Use FASTQ input for Tumor+UMI/Normal workflows. A fix is planned for a future release. | | **TMB (Tumor-Only)** | Tumor-only (T/O) TMB is less reliable than tumor/normal (T/N) TMB. Germline variants may bleed into the somatic variant count, inflating the TMB value. This limitation is documented in the User Guide and was present in prior releases. | For more accurate TMB reporting, use the tumor/normal workflow when a matched normal is available. | | **Hash Table Builder** | Building hash tables for some references (for example hg19 / hs37d5) may intermittently fail with `std::bad_alloc`. | Re-run the hash-table build; retries typically succeed. | | **BCL Convert** | With hardware BCL decompression enabled, `--bcl-output-format demux-autodetect` or `demuxmap` can crash (assertion) when used with relatively low `--num-threads` (for example 16) such that HW decompressors are active. Software decompression of the same CBCLs succeeds. | Use `--bcl-use-hw=false` or `--bcl-num-hw-decompressors=0` for demux-autodetect / demuxmap steps, or increase threads so HW decompressors are not allocated. | | **Installer** | The on-premises installer may report incorrect available/required disk space (including an inflated “additional space needed” value) and block installation even when sufficient space exists after temporary extract cleanup. | Retry with `--target` pointing to a large staging filesystem (for example `/staging/tmp`), or free space on the extract volume. Improved messaging is planned for a future release. | *** ## SW Installation Procedure ### Prerequisites * DRAGEN v4.6 software supports on-premises **Phase 3** or **Phase 4** DRAGEN servers. * **Note:** For on-premises analyses, TruPath analysis requires at least Phase 4 DRAGEN server due to FPGA memory limitations. For reference, Phase 4 servers have a serial number beginning with the letters "AC". * **Note:** v4.6 supports upcoming Phase 5 DRAGEN servers. * Supported operating systems: **el8/9** (Oracle Linux 8/9, AlmaLinux 8/9, and other RHEL-compatible distributions). * Root (sudo) privileges are required for on-server installation. ### Download DRAGEN v4.6 server installers are available at: Download the `.run` installer appropriate for your platform: ``` dragen-4.6.2-12.multi.el8.x86_64.run dragen-4.6.2-12.multi.el9.x86_64.run ``` For **Software Mode** (CPU-only): download `dragen-softwaremode-...bin` from the Illumina BioInsight Platform, or see [DRAGEN Software Mode](#dragen-software-mode) and the user guide. ### Verify Installer Integrity Verify the integrity of the downloaded installer before proceeding (example for el8): ```bash sudo sh dragen-4.6.2-12.multi.el8.x86_64.run --check ``` ### Install DRAGEN Software DRAGEN v4.6 uses the multi-version installer. Multiple compatible versions may coexist on the same server. ```bash sudo sh dragen-4.6.2-12.multi.el8.x86_64.run ``` (Use the `el9` installer on Oracle Linux 9 / el9-compatible systems.) After installation, DRAGEN v4.6 is available at: ``` /opt/dragen/4.6.2/bin/dragen ``` To add DRAGEN v4.6 to your PATH for the current user, add the following to `~/.bashrc`: ```bash export PATH="/opt/dragen/4.6.2/bin:$PATH" ``` To view all installed DRAGEN versions on the server: ```bash /usr/bin/dragen_versions ``` > **Note:** Installing a multi-version package will remove any previously installed single-version DRAGEN package (v4.2 or older). Installing a new multi-version package will NOT remove existing multi-version packages. ### Download Resource Files (Hash Tables) DRAGEN v4.6 uses the **same resource files as DRAGEN v4.5** (Hash Tables v12 and associated collections). Existing hash tables from DRAGEN v4.4 or earlier are **not compatible** and must be replaced if you have not already upgraded for v4.5. Download the appropriate pre-built hash tables for your reference(s) from: Extract hash tables to your reference directory (e.g., `/data/reference/`): ```bash tar xzvf hg38-alt_masked.cnv.graph.hla.methyl_cg.rna-12-r6.0-1.tar.gz -C /data/reference/ ``` ### Run System Self-Test > **Not applicable to Software Mode.** After installation on DRAGEN server, verify that the system is functioning correctly: ```bash /opt/dragen/4.6.2/self_test/self_test.sh ``` The self-test takes approximately 12 minutes. A `PASS` result confirms that the DRAGEN server, FPGA, and software are operating correctly. If you experience a `FAIL` result after installation, contact Illumina Technical Support. ### Licensing DRAGEN v4.6 requires valid Illumina licenses. Licenses are verified on each run. * For network-connected FPGA servers: licenses are automatically verified via `https://license.dragen.illumina.com`. * For Software Mode: an Illumina BioInsight Platform API key is required (see [DRAGEN Software Mode](#dragen-software-mode)). * For dark-site (air-gapped) FPGA servers: a license quota file must be installed. See the DRAGEN User Guide licensing section for instructions. For license installation assistance, contact your Illumina sales representative or Illumina Technical Support. ### Additional Resources | Resource | URL | | -------------------------- | ------------------------------------------------------------------------------------------------------- | | DRAGEN User Guide | | | DRAGEN Downloads | | | DRAGEN Product Files | | | Server Site Prep Guide | | | Illumina Technical Support | | *** | Revision | Date | Description | | -------- | -------------- | --------------- | | 01 | September 2026 | Initial release | *** *This document is proprietary. © Illumina, Inc. All rights reserved.* --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://help.connected.illumina.com/dragen/dragen-v4.6/reference/release-notes-readme/dragen-crn-v4.6.2.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.