# Copy Number Variant Calling

The DRAGEN Copy Number Variant (CNV) Pipeline detects copy number aberrations and regions with loss of heterozygosity (LOH) from next-generation sequencing (NGS) data. It supports germline and somatic workflows for both whole-genome sequencing (WGS) and whole-exome sequencing (WES) in a single interface via the DRAGEN Host Software.

## Choose Your Workflow

Use the table below to identify the right pipeline for your use case and jump directly to its documentation.

| Sample type | Data type | Input samples  | Documentation                                                                                                                     |
| ----------- | --------- | -------------- | --------------------------------------------------------------------------------------------------------------------------------- |
| Germline    | WGS / WES | Single / Multi | [Germline CNV Calling](https://help.dragen.illumina.com/product-guides/dragen-v4.5/dragen-dna-pipeline/cnv-overview/cnv-germline) |
| Somatic     | WGS / WES | Single         | [Somatic CNV Calling](https://help.dragen.illumina.com/product-guides/dragen-v4.5/dragen-dna-pipeline/cnv-overview/cnv-somatic)   |

Example commandlines are provided under [DRAGEN Recipes](https://github.com/illumina-swi/dragen-docs/blob/release/4.5-prod/product-guides/dragen-v4.5/user-guide/dragen-recipes/README.md).

Visit [Reference](https://help.dragen.illumina.com/product-guides/dragen-v4.5/dragen-dna-pipeline/cnv-overview/cnv-reference) for more details on the CNV component.

## Before You Begin

Before running the CNV pipeline, ensure the following prerequisites are in place:

1. **CNV-enabled reference hashtable** — The hashtable must be built with `--ht-build-cnv-hashtable true`. This generates an additional k-mer uniqueness map used to correct mappability biases.
2. **Aligned BAM or CRAM input** — The pipeline accepts pre-aligned reads. If you are starting from FASTQ, first run map/align or use streaming alignments.
3. **Panel of normals (required for WES)** — WES normalization requires a panel of normals. WGS can use self-normalization instead.

## Pipeline Overview

The DRAGEN CNV Pipeline processes the input signal through the following stages:

**1. Target Counts** — Read counts and other signals are extracted from alignments and binned into target intervals.

![](https://25033470-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FG9szlFZupV6Q2DasL98y%2Fuploads%2Fgit-blob-b9f011edb411777e8603fb4dbccfdedc4ca191b9%2Fcnv-calling.SignalTargetCounts.png?alt=media)

**2. Normalization** — The case sample is normalized against a panel of normals or against the estimated normal ploidy. Systematic biases (e.g., GC bias) are corrected to amplify event-level signals.

![](https://25033470-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FG9szlFZupV6Q2DasL98y%2Fuploads%2Fgit-blob-1ed9998ea6b8062f6cdafa88458df5c92e6f3f5c%2Fcnv-calling.SignalNormalized.png?alt=media)

**3. Segmentation** — The normalized signal is segmented using one of the available segmentation algorithms.

![](https://25033470-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FG9szlFZupV6Q2DasL98y%2Fuploads%2Fgit-blob-f303c4ee46d0b8ca21974200f6093f32f00416b7%2Fcnv-calling.SignalSegments.png?alt=media)

**4. CNV Calling** — Events are called from the segments, scored, and emitted in the output VCF.

![](https://25033470-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FG9szlFZupV6Q2DasL98y%2Fuploads%2Fgit-blob-c7fe7ecefe956cb573102a5f97104328284c4ec9%2Fcnv-calling.SignalCalledEvents.png?alt=media)