copy-number-variation

Overview

copy-number-variation covers detecting deletions and duplications from genomic data: read-depth methods for WGS/WES, SNP-array approaches (BAF/LRR), segmentation algorithms, and interpreting CNVs for rare disease and cancer. CNVs are a major cause of Mendelian disease and a key somatic alteration class — but detection is noisier than SNV calling and interpretation leans heavily on population databases.

When to use

• Germline CNV calling from WES/WGS: read-depth methods (GATK gCNV, CNVnator, Manta).
• SNP-array CNV analysis: PennCNV, LRR/BAF interpretation.
• Somatic copy-number in cancer: FACETS, ASCAT (purity/ploidy-aware), GISTIC for recurrent regions.
• Segmentation: CBS, HMM approaches, tuning sensitivity.
• Interpretation: DGV, gnomAD-SV, ClinGen dosage sensitivity.
• Validation: qPCR, MLPA, microarray confirmation.

Core concepts

• Read-depth principle. More copies → more reads. Normalize for GC content, mappability, and library size; compare against a reference panel (batch-matched — CNV calling without a good reference panel is unreliable). WES is uneven by capture efficiency — expect noisier calls than WGS; intronic/intergenic CNVs are invisible to WES.
• WGS structural-variant callers. Manta, Delly, LUMPY, GRIDSS combine read-depth with discordant-pair and split-read evidence — more precise breakpoints than depth alone. No single caller finds everything; ensemble approaches (e.g. Parliament2) improve recall. Validate breakpoint precision claims — many are ±kb, not base-pair.
• SNP arrays: LRR and BAF. Log R Ratio (total intensity → copy number) + B allele frequency (allelic ratio → LOH, mosaicism). PennCNV (HMM on LRR/BAF) remains standard. Arrays detect mosaicism and UPD that sequencing misses at standard depth — don't discard arrays as obsolete.
• Segmentation. Circular binary segmentation (DNAcopy), HMMs: partition the genome into constant-copy segments. Tuning sensitivity/specificity trade-off: permissive segmentation finds small events with more false positives. Require ≥3-10 consecutive probes/bins (platform-dependent) to call.
• Somatic copy-number. Tumor purity and ploidy confound everything: a deletion in 30% pure tumor looks like noise. FACETS/ASCAT jointly estimate purity, ploidy, and allele-specific copy number from tumor-normal pairs. GISTIC2 identifies significantly recurrent regions across cohorts. Report purity/ploidy with every somatic CNV call.
• Interpretation: population data. DGV and gnomAD-SV catalog benign CNVs — a "rare" deletion that's common in gnomAD-SV is not pathogenic. Check frequency by ancestry; small CNVs (<10 kb) are abundant in healthy genomes.
• Dosage sensitivity. ClinGen dosage-sensitivity map: haploinsufficiency (HI) and triplosensitivity (TS) scores per gene/region. A deletion's pathogenicity depends on whether it hits dosage-sensitive genes — a 2 Mb deletion of gene desert differs from 200 kb deleting a HI gene.
• De novo and inheritance. Trio analysis: de novo CNVs are strong candidates in rare disease; inherited CNVs need segregation and population-frequency assessment. Mosaic CNVs (detectable by BAF deviation or read-depth mixtures) explain some "negative" cases.
• Validation. qPCR/MLPA for small events, microarray for larger, FISH for mosaicism questions. NGS-only CNV calls, especially from WES, should be orthogonally confirmed before clinical reporting.

Practical workflow

1. Choose method. WGS for comprehensive SVs; WES for coding CNVs (with caveats); arrays for mosaicism/UPD or cost.
2. Build reference. Batch-matched normals for read-depth calling (≥30 samples ideal).
3. Call. GATK gCNV / Manta / PennCNV as appropriate; ensemble for WGS SVs.
4. Filter. Size, probe/bin support, population frequency (gnomAD-SV, DGV), quality scores.
5. Interpret. Gene content, ClinGen HI/TS scores, inheritance (trio), phenotype match.
6. Validate. qPCR/MLPA/array confirmation of candidates.
7. Report. Coordinates (with build), size, gene content, classification rationale, validation status.

Example command sketch:

# GATK gCNV (cohort mode needs a panel of normals built first)
gatk GermlineCNVCaller --run-mode COHORT -L intervals.list \
  --contig-ploidy-calls ploidy-calls --output cnv_cohort/

Common pitfalls

• WES CNV calls without batch-matched reference panels.
• Tumor CNVs without purity/ploidy estimation.
• Common gnomAD-SV variants reported as pathogenic.
• Gene-desert CNVs over-interpreted; HI-gene deletions missed.
• Breakpoint precision overstated (kb-level presented as exact).
• Mosaicism missed (no BAF analysis).
• NGS-only calls clinically reported without orthogonal validation.