Copy Number Variation Analysis — CNV Detection and Interpretation

Variant calling (SNP/indel calling) detects single-nucleotide changes and small insertions or deletions, typically up to a few dozen base pairs. CNV analysis detects larger structural changes — gains or losses of kilobase-to-megabase segments — by measuring read depth or hybridisation intensity rather than the exact sequence. A complete genomic characterisation of a tumour or rare-disease sample usually requires both analyses applied independently and then integrated.

SNP arrays are cost-effective, well-standardised, and provide allele frequency information useful for LOH detection; they are the standard for clinical germline CNV testing and large population studies. WGS provides higher resolution (detecting smaller events), base-pair breakpoint precision, and simultaneous access to SNVs and indels, but at higher cost and more complex bioinformatics. The choice depends on budget, required resolution, and whether simultaneous variant calling is needed.

Low purity (below ~30%) severely dampens the log2 ratio signal, causing true deletions to appear near diploid. Use purity- and ploidy-aware callers (PURPLE, ABSOLUTE, or ASCAT) that jointly estimate the tumour cell fraction and genome ploidy, then adjust copy number thresholds accordingly. Increasing sequencing depth and using allele-specific information (B-allele frequency) from heterozygous SNPs substantially improves sensitivity in low-purity samples.

The optimal penalty depends on expected CNV size and noise level. For cancer genomes with focal amplifications, a finer segmentation (lower penalty) is appropriate; for germline studies or low-coverage data, a coarser segmentation reduces false positives. A practical approach is to test two or three penalty values, visualise the resulting segments against the raw coverage track, and choose the one that captures visually obvious transitions without fragmenting flat regions. Cross-validate top events with a second caller or orthogonal assay.

RNA-seq-based CNV inference is possible — tools such as InferCNV exploit systematic shifts in normalised gene expression across genomic regions to infer copy number, particularly in single-cell RNA-seq tumour data. However, expression is driven by many factors beyond copy number (transcription regulation, splicing, degradation), so RNA-based CNV calls are less reliable than DNA-based ones. Use RNA-seq CNV inference only when DNA is unavailable, and treat results as indicative rather than definitive.