Genomics & Variant Analysis

CNV and structural variation analysis: multi-caller copy-number segments and breakpoint-resolved SV VCFs from WGS or WES BAMs

Copy-number and structural variation (CNV/SV) analysis detects genome-wide dosage changes and rearrangements from read depth and paired-end or split-read evidence [1]. Pepkio returns version-pinned segment tables, SV VCFs, gene-overlap annotations, and scripts, with custom and bespoke workflow support, for academic, biotech, and pharma teams starting from BAMs or scoped alignments. Collins et al. [1] reported a median of 7,439 high-quality SVs per genome in gnomAD-SV. That figure is a reference benchmark, not a Pepkio deliverable.

Key facts

Key facts about CNV and Structural Variation
FactValue
Supported platforms / instrumentsPre-aligned BAMs or CRAMs from Illumina NovaSeq X / 6000 / NextSeq 2000, HiSeq 2500/4000; MGI DNBSEQ (BGI) short-read WGS/WES; Element Biosciences AVITI and Ultima Genomics UG100 when scoped at kickoff. WGS preferred; WES capture-aware. Re-alignment from FASTQ scoped separately (BWA-MEM2 2.2.1)
Input requirementsWGS: ≥30× mean autosomal depth and >95% callability recommended [7]. WES: ≥50-100× on-target depth; all callers perform poorly on sparse WES relative to WGS [4]. gCNV cohort model: ≥100 technically matched WGS samples [3]. Smaller cohorts or single-sample: CNVkit 0.9.11 or GATK-SV single-sample mode with a pre-computed reference panel [3]
Reference builds supportedHuman GRCh38 primary (GATK resource bundle 4.6); legacy GRCh37/hg19 on request; mouse GRCm39; custom references scoped at kickoff
Primary tools (with versions)GATK 4.6.0.0 (GermlineCNVCaller / gCNV); Manta 1.6.0; DELLY 1.2.6; Lumpy 0.2.13 via smoove 0.2.8; CNVkit 0.9.11 when scoped; cn.MOPS 1.56.0 for cohort depth CNVs; samtools/bcftools 1.21; mosdepth 0.3.3; bedtools 2.31.1; Picard 3.2.0; MultiQC 1.25
Typical turnaround time2-4 weeks (single-sample from BAM); 4-8 weeks (multi-sample cohort with gCNV model training and multi-caller merge). Confirmed at kickoff
Deliverable formats.cnv.vcf.gz, .sv.vcf.gz, segment .bed/.tsv; PDF/SVG figures; HTML QC report; documented bash/R/Python scripts; Methods draft
Key cited best-practice referenceGabrielaite et al. [4], Cancers; Collins et al. [1], Nature; Mahmoud et al. [2], Genome Biology
Custom / bespoke analysisNon-standard inputs, outputs, and methods scoped at kickoff, e.g. somatic tumor-normal CNV/SV, locus-specific re-calling, client reference panels, array validation overlays, CytoScan comparison tables, or non-standard export formats

What is CNV and structural variation analysis?

CNV and structural variation analysis segments read-depth profiles to call copy-number gains and losses, then uses discordant read pairs and split reads to detect rearrangements typically ≥50 bp [2, 3]. It measures dosage and large-scale architecture, not SNVs and indels, which need separate variant calling [4]. Pepkio returns annotated segment tables and SV VCFs from client BAMs or scoped alignments, with sensitivity limits documented per assay. See the CNV and structural variation glossary.

When should you use CNV and structural variation analysis?

CNV/SV analysis fits when SNV/indel calls are not enough: undetected dosage changes, translocations, inversions, or a negative WES/WGS SNV panel in rare disease, and when aligned WGS or WES data already exist or are planned.

Comparison of CNV/SV calling approaches
ApproachBest forLimitationsApproximate cost range
CNV/SV from WGS/WES BAMs (this service)Genome-wide dosage changes, breakpoint-resolved SVs, undiagnosed disease after negative SNV panelsWES misses many non-coding and large CNVs; short-read data often miss complex SVs and repeat-rich loci [4, 5]Bioinformatics-only; usually lower than a full FASTQ pipeline when BAMs exist
SNV/indel variant calling onlyBase-pair mutations, cohort SNV harmonizationDoes not reliably detect most CNVs or balanced SVsLower compute when CNV/SV is not required
SNP-array / CytoScan CNVClinical-grade CNV at established loci with array validation historyFixed probe coverage; limited breakpoint resolution and novel SV discoveryLowest genotyping cost; resolution limited by probe density
Long-read SV (separate spoke)Complex, repeat-rich, phased structural eventsHigher per-base sequencing cost than short-read WGSLibrary prep + sequencing + bioinformatics vary by platform and depth
  • Rare disease trios negative on WES: Gabrielaite et al. [4] benchmarked 11 CNV callers on matched WES and WGS; WGS outperformed WES for CNV recall, and no single caller captured all events. They recommend combining GATK gCNV, Lumpy, DELLY, and cn.MOPS.
  • Population-scale SV reference catalogs: Collins et al. [1] integrated Manta, Wham, gCNV, and cn.MOPS across 14,891 genomes in gnomAD-SV, producing a harmonized SV reference for allele-frequency filtering.
  • Complex de novo SVs in undiagnosed disease: Jung et al. [6] analyzed 13,698 probands from the UK 100,000 Genomes Project; 8.4% of de novo SVs were complex, and 22% of array- or WES-called simple deletions/duplications were reclassified as complex on WGS.

How the analysis works — step by step

  1. 1. Validate BAM/CRAM inputs and metadata

    Pepkio verifies MD5 checksums, coordinate-sorted BAM or CRAM structure, indexed .bai/.crai presence, and @RG tags. Reference build, sample sex, assay type, platform, and batch are recorded in sample_manifest.csv. Mismatched builds or missing indices are flagged before calling.

    Tools and outputs

    Tools used: md5sum; samtools 1.21 quickcheck; custom validation scripts

    Output: sample_manifest.csv with sample IDs, reference build, assay type, platform, input path, and QC flags

  2. 2. Audit coverage and callability

    mosdepth and Picard CollectWgsMetrics or CollectHsMetrics report mean coverage, pct at ≥15×/≥20×, duplicate rate, and insert-size distribution. Sub-threshold samples are flagged [7, 8].

    Tools and outputs

    Tools used: mosdepth 0.3.3; Picard 3.2.0 CollectWgsMetrics or CollectHsMetrics

    Output: coverage_summary.csv; mosdepth.global.dist.txt; wgs_metrics.txt or hs_metrics.txt; coverage histogram and CDF plots

  3. 3. Prepare bin-level read-depth profiles

    Genomic bins match the assay: uniform bins for WGS (GATK defaults start at ~1 kb; larger when scoped) and capture-aware intervals for WES [9]. GC bias and mappability are assessed per bin; blacklisted regions and sex chromosomes are handled per project design.

    Tools and outputs

    Tools used: bedtools 2.31.1; mosdepth 0.3.3; custom bin-generation scripts

    Output: {project}.wgs_bins.bed or {project}.capture_targets.bed; depth_by_bin.tsv; GC bias diagnostic plots

  4. 4. Train or select the CNV reference model

    For cohorts with ≥100 matched germline WGS samples, GATK GermlineCNVCaller trains a cohort gCNV model [3]. Smaller cohorts, somatic projects, or WES use CNVkit or GATK-SV single-sample mode with a pre-computed reference panel [3]. Training samples should match cases in library prep and batch [1].

    Tools and outputs

    Tools used: GATK 4.6.0.0 GermlineCNVCaller model training; CNVkit 0.9.11 batch reference build when scoped

    Output: {batch}.gcnv_model/ or {project}.cnvkit_reference.cnn; model training QC report; cnv_reference_manifest.csv

  5. 5. Call copy-number variants

    GATK gCNV infers copy-number states from cohort-trained depth models; CNVkit applies GC correction and CBS segmentation [9]; cn.MOPS models read counts across cohort samples [10]. Per-sample CNV counts are checked before merging.

    Tools and outputs

    Tools used: GATK 4.6.0.0 PostprocessGermlineCNVCalls; CNVkit 0.9.11 batch / call; cn.MOPS 1.56.0 when scoped

    Output: {sample}.gcnv_calls.vcf.gz; {sample}.cnvkit.cns; {sample}.cnmops_segments.bed; per-sample CNV count summary

  6. 6. Call structural variants from read signatures

    Manta and DELLY call SVs from discordant pairs, split reads, and depth evidence by default; Lumpy (via smoove) is added when scoped [11, 12, 13]. Per-caller VCFs retain QUAL, FILTER, and evidence tags for merge decisions.

    Tools and outputs

    Tools used: Manta 1.6.0; DELLY 1.2.6; smoove 0.2.8 (Lumpy 0.2.13) when scoped

    Output: {sample}.manta.sv.vcf.gz; {sample}.delly.sv.vcf.gz; {sample}.lumpy.sv.vcf.gz when scoped; per-caller variant count summaries

  7. 7. Merge and filter multi-caller evidence

    CNV segments and SV calls are merged using reciprocal overlap thresholds agreed at kickoff [4]. Size, quality, and bin-count filters are tuned per assay [9].

    Tools and outputs

    Tools used: bcftools 1.21; bedtools 2.31.1; SURVIVOR 1.0.6 or custom merge scripts when scoped

    Output: {cohort}.merged_cnv.bed; {cohort}.merged_sv.vcf.gz; cnv_sv_filter_summary.csv; caller overlap Venn data

  8. 8. Annotate CNV/SV with genes and functional context

    CNV segments and SV breakpoints are intersected with Ensembl/GENCODE gene models. gnomAD-SV allele frequencies and ClinVar overlap are added when reference databases are configured [1].

    Tools and outputs

    Tools used: bedtools 2.31.1; bcftools 1.21; Ensembl BioMart or custom annotation scripts

    Output: cnv_gene_overlap.tsv; sv_annotation_master.tsv; annotated .vcf.gz when requested

  9. 9. Run CNV/SV sanity QC

    Per-sample CNV burden, SV type breakdown, and size distributions are compared across the cohort. Outlier call counts or batch clustering trigger review before delivery [4]. CytoScan or MLPA truth sets are overlaid when clients provide coordinates.

    Tools and outputs

    Tools used: custom R/Python QC scripts; bcftools 1.21 stats

    Output: cnv_sv_qc_summary.csv; CNV burden and SV type bar charts; batch stratification report

  10. 10. Aggregate QC and package deliverables

    MultiQC aggregates coverage, CNV, and SV metrics across samples. Final scripts, README, Methods draft, and HTML QC report are packaged per agreed retention policy.

    Tools and outputs

    Tools used: MultiQC 1.25

    Output: MultiQC report; final deliverable bundle with scripts, figures, and Methods draft

What Pepkio delivers

Processed data files

  • Per-sample CNV .bed/.vcf.gz; per-caller SV VCFs (Manta and DELLY default; Lumpy when scoped)
  • Merged cohort outputs when in scope
  • Annotation and QC tables: cnv_gene_overlap.tsv, sv_annotation_master.tsv, coverage_summary.csv, cnv_sv_qc_summary.csv, cnv_sv_filter_summary.csv

Figures (PDF/SVG)

  • CNV profile; segment size histogram; CNV burden bar chart
  • SV type breakdown; caller overlap summary; locus zoom plots when scoped

Tables

  • cnv_gene_overlap.tsv (chrom, start, end, CN, genes)
  • sv_annotation_master.tsv (SVTYPE, SVLEN, genes, gnomAD-SV AF when configured)
  • cnv_sv_qc_summary.csv (depth, CNV/SV counts, batch ID)

Code

  • Commented bash, R, and Python scripts; environment lock files
  • Private Git or agreed file transfer

Documentation

  • QC report; README; Methods draft
  • Post-delivery reviewer support within agreed scope

Technical decisions we make — and why

Multi-caller default: gCNV + Manta + DELLY (+ Lumpy when scoped)
Gabrielaite et al. [4] recommend combining GATK gCNV, Lumpy, DELLY, and cn.MOPS because no single CNV caller captures all variant classes reliably. Single-caller delivery is available when scoped.
gCNV cohort model vs CNVkit
GATK GermlineCNVCaller trains a cohort model when ≥100 matched germline WGS samples are available [3]. CNVkit 0.9.11 is the default for smaller cohorts, somatic designs, and WES [4, 9].
WGS vs WES expectations
WGS is the default for genome-wide CNV and SV discovery [4]. WES CNV/SV is scoped with capture-kit-aware bins and documented sensitivity limits outside capture targets.
Filter stringency: project-tuned, not universal cutoffs
Filters depend on assay, cohort size, and use case. Pepkio documents thresholds per project rather than applying generic SNV cutoffs [9, 14].
Batch handling for gCNV and depth models
GATK-SV recommends separate gCNV models per batch when library protocols differ [3]. Pepkio stratifies batch in QC and trains batch-specific models when required.

Common questions

What is the minimum sequencing depth for reliable CNV and SV calling?

For germline CNV/SV from WGS BAMs, Pepkio recommends ~30× mean autosomal depth with >95% callability [7]. WES requires ≥50-100× on-target depth [4]. Sarwal et al. [15] found Manta and LUMPY among the highest F-scores for deletion calling. Thresholds are confirmed at kickoff.

Can you call CNVs from my existing WES BAMs?

Yes, when on-target depth and capture-kit metadata are documented. WES misses many off-target events that WGS detects [4]. Pepkio applies capture-aware CNVkit binning when scoped and documents sensitivity limits in the Methods draft.

Can you analyze low-quality or low-coverage BAMs?

Sometimes, with caveats. Mean depth below ~20× on WGS or sparse on-target coverage on WES can reduce CNV segment confidence and SV evidence counts [7]. Sub-threshold samples are flagged before calling; re-sequencing is discussed when feasible. Partial calling on agreed chromosomal regions or gene lists is possible when full-genome CNV/SV is not justified.

Do you support BAMs from Illumina, MGI DNBSEQ, Element AVITI, and Ultima UG100?

Pepkio processes pre-aligned BAMs from Illumina and MGI DNBSEQ (BGI) platforms when @RG tags and reference build are documented. Element AVITI and Ultima UG100 BAMs are accepted when scoped at kickoff. If only FASTQs are available, re-alignment with BWA-MEM2 2.2.1 is scoped separately or via the WGS/WES spokes.

How long does CNV and structural variation analysis take at Pepkio?

Single-sample CNV/SV calling from client BAMs typically completes in 2-4 weeks; multi-sample cohorts with gCNV training and merge typically take 4-8 weeks. Timelines depend on sample count, assay, and scope, and are confirmed at kickoff.

How do you handle batch effects in multi-batch CNV/SV cohorts?

CNV depth models are sensitive to library prep, sequencer, and batch-specific coverage biases [1, 3]. Pepkio trains separate gCNV models per batch when dosage scores differ, stratifies batch in QC reports, and flags batch-clustered samples. Batch-aware CNVkit reference panels are constructed from matched normals when available.

Do you use one caller or combine Manta, DELLY, and gCNV?

Pepkio defaults to GATK gCNV (or CNVkit) plus Manta and DELLY because no single tool captures all CNV classes reliably [4]. Lumpy and cn.MOPS are added when scoped. Single-caller workflows are available with documented trade-offs.

How do you handle complex structural variants?

Short-read pipelines detect many complex SVs but cannot resolve all breakpoint architectures that long-read sequencing can resolve [5, 6]. Pepkio reports complex SVTYPE calls when assigned; unresolved events are flagged for long-read follow-up when scoped.

Do I own the code, and in what format is it delivered?

Yes. You retain full ownership of code, scripts, and results. Pepkio delivers commented bash, R, and Python scripts with environment lock files. Jupyter or R Markdown delivery is available on request.

Can I be involved during analysis?

Yes. Checkpoint reviews occur after coverage audit, post-calling sanity checks, and before final delivery. A PhD-level scientific contact leads the project and documents decisions at each stage.

What does post-delivery reviewer support include?

Clarification of methods, merge thresholds, QC metrics, and minor figure or table revisions within agreed scope. Methods and Supplementary drafts are included for analyses we performed; substantial new reviewer requests are scoped separately.

Is co-authorship required?

No. Pepkio does not require co-authorship unless that is discussed explicitly. Acknowledgment of bioinformatics support is standard practice.

Related services

  • Whole-genome sequencing — Full FASTQ-to-BAM pipeline when alignment and raw-read QC are needed before CNV/SV calling.
  • Whole-exome sequencing — Capture-enriched libraries with on-target depth QC and coding-region variant discovery.
  • Variant calling — SNV/indel calling from the same BAMs to build a complete small-variant catalog alongside CNV/SV.
  • Long-read DNA sequencing — Complex, repeat-rich, and phased structural events that short-read callers often cannot fully resolve.
  • Custom consulting — Pre-project depth, cohort size, caller strategy, and array-validation planning before sequencing or CNV/SV analysis.
References
  1. Collins RL, Brand H, Karczewski KJ, et al. A structural variation reference for medical and population genetics. Nature. 2020;581(7809):444–451. https://doi.org/10.1038/s41586-020-2287-8 (PMID: 32461652)
  2. Mahmoud M, Gobet N, Cruz-Dávalos DI, et al. Structural variant calling: the long and the short of it. Genome Biology. 2019;20:246. https://doi.org/10.1186/s13059-019-1828-7 (PMID: 31747936)
  3. Broad Institute. GATK-SV structural variation discovery documentation. 2024. https://gatk.broadinstitute.org/hc/en-us/articles/9022487952155-Structural-variant-SV-discovery; https://github.com/broadinstitute/gatk-sv
  4. Gabrielaite M, Torp MH, Rasmussen MS, et al. A comparison of tools for copy-number variation detection in germline whole exome and whole genome sequencing data. Cancers. 2021;13(24):6283. https://doi.org/10.3390/cancers13246283 (PMID: 34944901)
  5. Ebert P, Audano PA, Zhu Q, et al. Haplotype-resolved diverse human genomes and integrated analysis of structural variation. Science. 2021;372(6537):eabf7117. https://doi.org/10.1126/science.abf7117 (PMID: 33632895)
  6. Jung H, Yang TP, Walker S, et al. Complex de novo structural variants are an underestimated cause of rare disorders. Nature Communications. 2025;16:9528. https://doi.org/10.1038/s41467-025-64722-2 (PMID: 41184278)
  7. Rehm HL, Bale SJ, Bayrak-Toydemir P, et al. Best practices for the analytical validation of clinical whole-genome sequencing intended for the diagnosis of germline disease. npj Genomic Medicine. 2021;6(1):47. https://doi.org/10.1038/s41525-020-00154-9 (PMID: 33110627)
  8. Ajay SS, Parker SCJ, Abaan HO, et al. Accurate and comprehensive sequencing of personal genomes. Genome Research. 2011;21(9):1498–1505. https://doi.org/10.1101/gr.123638.111 (PMID: 21771779)
  9. Talevich E, Shain AH, Botton T, Bastian BC. CNVkit: genome-wide copy number detection and visualization from targeted DNA sequencing. PLoS Computational Biology. 2016;12(4):e1004873. https://doi.org/10.1371/journal.pcbi.1004873 (PMID: 27100738)
  10. Klambauer G, Schwarzbauer K, Mayr A, et al. cn.MOPS: mixture of Poissons for discovering copy number variations in next-generation sequencing data with a low false discovery rate. Nucleic Acids Research. 2012;40(9):e69. https://doi.org/10.1093/nar/gks003 (PMID: 22302147)
  11. Chen X, Schulz-Trieglaff O, Shaw R, et al. Manta: rapid detection of structural variants and indels for germline and cancer sequencing applications. Bioinformatics. 2016;32(8):1220–1222. https://doi.org/10.1093/bioinformatics/btv710 (PMID: 26647377)
  12. Rausch T, Zichner T, Klipp F, et al. DELLY: structural variant discovery by integrated paired-end and split-read analysis. Bioinformatics. 2012;28(18):i333–i339. https://doi.org/10.1093/bioinformatics/bts378 (PMID: 22962449)
  13. Layer RM, Chiang C, Quinlan AR, Hall IM. LUMPY: a probabilistic framework for structural variant discovery. Genome Biology. 2014;15(6):R84. https://doi.org/10.1186/gb-2014-15-6-r84 (PMID: 24970577)
  14. Koboldt DC. Best practices for variant calling in clinical sequencing. Genome Medicine. 2020;12:91. https://doi.org/10.1186/s13073-020-00791-w (PMID: 33106175)
  15. Sarwal V, Niehus S, Ayyala R, et al. A comprehensive benchmarking of WGS-based deletion structural variant callers. Briefings in Bioinformatics. 2022;23(5):bbac221. https://doi.org/10.1093/bib/bbac221 (PMID: 35753701)

Let's Talk About Your Science

Tell us:

  • • Your biological question
  • • Data type and size
  • • Timeline constraints

We'll tell you:

  • • What's feasible
  • • How long it will take
  • • Exactly what it will cost
Discuss Your Project

Contact us to start with a free consultation. Need everyday bench calculators? Try our free lab tools.