cnv-caller-agent

---

name: 'cnv-caller-agent' description: 'AI-enhanced copy number variation calling and analysis from sequencing data for cancer genomics, constitutional CNV detection, and chromosomal aberration characterization.' measurable_outcome: Execute skill workflow successfully with valid output within 15 minutes. allowed-tools:

• read_file
• run_shell_command

---

CNV Caller Agent

The CNV Caller Agent provides comprehensive AI-enhanced copy number variation analysis from WGS, WES, and targeted sequencing for cancer genomics and constitutional CNV detection.

When to Use This Skill

• When calling somatic CNVs from tumor-normal paired sequencing.
• To detect constitutional CNVs from germline sequencing.
• For allele-specific copy number analysis.
• When characterizing focal amplifications and deletions in cancer.
• To assess tumor purity and ploidy from CNV data.

Core Capabilities

1. Somatic CNV Calling: Detect tumor-specific copy number alterations.

2. Germline CNV Detection: Identify constitutional CNVs for rare disease.

3. Allele-Specific Analysis: Determine allele-specific copy number and LOH.

4. Purity/Ploidy Estimation: Estimate tumor content and genome doubling.

5. Focal Event Detection: Identify amplifications and deletions of driver genes.

6. Segmentation Optimization: AI-enhanced breakpoint detection.

Workflow

1. Input: BAM files (tumor/normal), or targeted panel data.

2. Coverage Normalization: GC correction, mappability adjustment.

3. Segmentation: Identify regions of consistent copy number.

4. Allele-Specific: Calculate B-allele frequency for heterozygosity.

5. Purity/Ploidy: Estimate sample parameters.

6. Calling: Assign integer copy number states.

7. Output: Segmented CNV calls, purity/ploidy, driver events.

Example Usage

User: "Call somatic copy number alterations from this tumor-normal WES pair."

Agent Action:

python3 Skills/Genomics/CNV_Caller_Agent/cnv_caller.py \
    --tumor tumor.bam \
    --normal normal.bam \
    --reference GRCh38.fa \
    --method facets \
    --targets exome_targets.bed \
    --driver_genes cancer_genes.txt \
    --output cnv_results/

CNV Calling Methods

Tool	Application	Key Features
FACETS	Tumor WES	Purity/ploidy, allele-specific
ASCAT	Tumor WGS/arrays	Allele-specific, multi-clone
CNVkit	WES/targeted	Hybrid reference approach
GATK CNV	WES/WGS	GATK ecosystem integration
Purple	WGS	GRIDSS integration, comprehensive
CONICS	scRNA-seq	Single-cell CNV inference

Key Output Metrics

Metric	Description	Interpretation
Purity	Tumor fraction	Sample quality
Ploidy	Average copy number	Genome doubling
LOH	Loss of heterozygosity	Regions of allele loss
SCNA burden	Total altered fraction	Genomic instability
Focal events	Amplifications/deletions	Driver candidates

Cancer Driver CNVs

Gene	Alteration	Cancer Type
ERBB2 (HER2)	Amplification	Breast, gastric
MYC	Amplification	Many cancers
EGFR	Amplification	Lung, GBM
CDK4/MDM2	Amplification	Sarcoma, GBM
CDKN2A	Deletion	Many cancers
RB1	Deletion	Many cancers
PTEN	Deletion	Prostate, GBM

AI/ML Enhancements

Segmentation:

• Deep learning for breakpoint detection
• Noise reduction in low-coverage data
• Improved sensitivity for focal events

Quality Prediction:

• Sample quality scoring
• Artifact detection
• Confidence estimation

Driver Prioritization:

• GISTIC-style analysis
• Functional impact scoring
• Pan-cancer frequency context

Allele-Specific Copy Number

Total CN = Major allele + Minor allele

Examples:
- Normal: 1 + 1 = 2 (diploid)
- CN gain: 2 + 1 = 3 (trisomy)
- CN-LOH: 2 + 0 = 2 (normal total, LOH)
- Homozygous deletion: 0 + 0 = 0
- High amplification: 10 + 0 = 10 (focal amp)

Prerequisites

• Python 3.10+
• CNV calling tools (FACETS, CNVkit, etc.)
• Reference genome and annotations
• Sufficient memory for WGS (16GB+)

Related Skills

• Variant_Interpretation - For CNV annotation
• HRD_Analysis_Agent - For HRD scoring from CNV
• Pan_Cancer_MultiOmics_Agent - For pan-cancer CNV context

Quality Considerations

1. Coverage depth: Higher = better resolution
2. Tumor purity: Low purity challenges calling
3. Normal match: Best with matched normal
4. Target design: Uniform coverage for panels
5. GC bias: Proper normalization critical

Author

AI Group - Biomedical AI Platform