An integrated platform for concurrent structural and single-nucleotide variants improves copy-number detection and reveals pathogenic alleles in undiagnosed Mendelian families Authors: Haowei Du, Ming Yin Lun, Lidiia Gagarina, Jesse D. Bengtsson, Christopher M. Grochowski, Michele G. Mehaffey, James Paul Hwang, Shalini N. Jhangiani, Sravya V. Bhamidipati, Donna M. Muzny, M. Cecilia Poli, Sebastian Ochoa, Ivan K. Chinn, Anna Lindstrand, Jennifer E. Posey, Richard A. Gibbs, Pengfei Liu, James R. Lupski, Claudia M. B. Carvalho Published: 2025 ABSTRACT BackgroundCopy number variation (CNV) is a class of genomic structural variation (SV) that contributes to genomic disorders and can significantly impact health. Short-read genome sequencing (sr-GS) enables genome-wide SV calling which has been shown to increase diagnosis in unsolved rare disease families. The growing number of large sequencing cohort projects with sr-GS data available requires open free analytical tools that provide visualization of CNV and SV integrated calls associated with gene annotation, proband-parent trio analysis to enable prioritization of de novo variants, B-allele frequency (BAF) plots to support CNV calls, parent of origin assessment and mosaicism detection.MethodsTo support those needs, we developed VizCNV, an open-source platform that incorporates read depth and BAF to enable haplotype-aware CNV analysis. The tool incorporates multiple interactive view modes for SV concurrent calls and annotation tracks for analyzing chromosomal abnormalities [e.g., aneuploidy, segmental aneusomy, and chromosome translocations], gene exonic rearrangements and non-coding gene regulatory regions. In addition, VizCNV includes a built-in filter schema for trio genomes, prioritizing the detection of de novo CNVs. We optimized VizCNV using 1000 Genomes Project data and benchmarked its performance against a cohort containing CNVs validated by multiple technologies. Finally, we applied VizCNV to a molecularly unsolved primary immunodeficiency disease cohort (PIDD, n = 39) previously analyzed by exome sequencing.ResultsUpon computational optimization, VizCNV achieved approximately 82.3% recall and 76.3% precision for deletions > 10 kb. VizCNV accurately detected all 71 validated copy number gains and correctly indicated potential underlying genomic complexities. Haplotype-aware CNV analysis identified a meiosis I non-disjunction event (trisomy 21), three de novo CNVs at two unique loci and 48 inherited candidate CNVs in the PIDD cohort of which 42% (20/48) were validated by integrated CNV/BAF analysis. Moreover, genotype–phenotype analyses revealed that a compound heterozygous combination of a paternal 12.8 kb deletion of exon 5 and a maternal missense variant allele of DOCK8 are the molecular cause of one proband diagnosed with Hyper-IgE syndrome.ConclusionsVizCNV provides a robust and flexible platform for identification of aneuploidies, CNV, SV discovery and visualization of CNV and BAF data. It is also a useful tool to investigate features of genomic rearrangements such as parental origin which has implications for genetic counseling and mechanistic studies. The tool is freely available through https://doi.org/10.6084/m9.figshare.25869523.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13073-025-01593-8. Background Copy number variation (CNV), a class of structural variation (SV) that results in deviations from the diploid state, represents a major source of genetic diversity, has significant implications for human health, and contributes to a variety of genomic disorders [1]. These structural genomic alterations range from large-scale changes like whole chromosome duplications (e.g., trisomies) to smaller CNVs that may affect as little as a single exon. Pathogenic CNVs change the expression of dosage-sensitive genes at loci for autosomal dominant traits and contribute to biallelic variation at loci for autosomal recessive traits. Over the past decades, CNV detection has progressed from low-resolution karyotyping of large chromosomal gains and losses to high-throughput moderate resolution methods – most notably array comparative genomic hybridization (aCGH) and genome-wide single nucleotide polymorphism (SNP) arrays. Modern next-generation, massively parallel DNA sequencing, including exome sequencing (ES) and genome sequencing [GS; both short read (sr) and long read (lr)-GS], provides even finer resolution. However, CNV calling is challenged by a high number of false positive calls, particularly for copy number gains and those smaller than 50 kb. This problem may be amplified by the overwhelming number of SV calls generated by SV callers mostly based on the presence of split-reads or deviation of expected genome distances and orientation of read pairs [2]. The integration of read depth, B-allele frequency (BAF) information, and haplotype phasing to the CNV analysis has been pivotal in high-impact variant prioritization [3, 4], and enables the identification of: i) de novo CNVs – i.e. mutations present only in the proband, but not in either parent [5], ii) homozygous deletions [6] and duplications [7], iii) variants rendered homozygous by identity by descent [8], and uniparental disomy [9, 10]. Such integrative approaches have shown great potential in narrowing the search space for pathogenic variant screening, identifying novel candidate genes in rare diseases, and uncovering novel disease mechanisms [11, 12]. Methods for concurrent whole genome haplotyping and copy number profiling through an array-based platform have shown great utility in prenatal testing and molecular diagnosis of genomic disorders [13, 14]. The resolution and sensitivity of haplotyping with SNP arrays are, however, region-specific, and depends on the design of the probes and density of SNPs. A mother-father-proband trio GS dataset has the potential to enhance the resolution of CNV haplotyping by enabling the calling of both common and rare SNVs within a family. The numerous bioinformatic steps involved increase the complexity of the analysis, hindering its widespread application to GS datasets. To address these challenges, we developed VizCNV, a platform specifically designed for the concurrent BAF, CNV and SV analysis on GS data in the context of rare genetic disease research. VizCNV is shown to enhance the application of sequencing data by incorporating BAF, SV information, and parental origin to streamline the process of identifying, and prioritizing high-impact allele-specific CNVs, including de novo variants. The significance of being able to detect CNVs based on read depth information, independently from split-reads and discordant read-pair-based SV callers, is showcased by the analysis of a cohort of probands with known pathogenic CNVs. Additionally, the utility of a combined CNV/SV/BAF analysis to uncover pathogenic variants is highlighted in probands diagnosed with Primary Immunodeficiency Disease (PIDD), and who were molecularly unsolved by ES. VizCNV is an open-resource software built on the R Shiny framework, designed to be interactive, flexible, and user-friendly. Methods Read depth normalization and segmentation A BED (Browser Extensible Data) file that stores the genomic position and sequencing coverage measured from a BAM (Binary Alignment Map) or CRAM (Compressed Reference-oriented Alignment Map) file is required. While multiple tools can generate such a file, VizCNV seamlessly supports the BED file output from Mosdepth [15]. Two choices are available for normalization and segmentation, respectively. By default, the read depth chromosomal median is used to calculate the log2 ratio for every 1000 bp. Additionally, there is an option to use the genome-wide median, which facilitates chromosomal analysis, such as detecting aneuploidy, for read depth normalization. The log2 ratio data points can be segmented using the Circular Binary Segmentation (CBS) algorithm [16] implemented in the DNAcopy Bioconductor package or shift level model implemented in the SLMSuite R package [17]. Problematic regions of the genome (build GRCh38 and GRCh37/hg19), as defined by those non-uniquely specified DNA stretches outside the ‘assayable portion’ of the human genome [18], are masked using University of California Santa Cruz (UCSC) Genome Browser Database RepeatMasker and segmental duplication (SD) tracks (https://genome.ucsc.edu/cgi-bin/hgTables?). Trio genome-wide SNV genotyping and phased BAF segmentation Germline variants per sample, including SNVs and small insertions and deletions (INDELs), were called using the Genome Analysis Toolkit (GATK) HaplotypeCaller in ‘-ERC GVCF’ mode. This generated a genomic VCF (gVCF) file for each family member: the proband, mother, and father. Subsequently, joint genotyping was conducted on the combined gVCFs for the entire family, facilitating genotyping at all variant positions. Variants were excluded from phasing if they met any of the following criteria, i) absence of genotype information for any family member, ii) low coverage depth in any one family member (read depth < 8), and iii) presence of an INDEL or being a multiallelic site [19]. SNVs were phased confined to Mendel’s expectations. The variant read to total read (VR/TR) ratio of phased B-allele were segmented using the CBS algorithm [16]. Since only informative heterozygous SNVs were able to be phased and segmented, regions with runs of homozygosity (ROH) will be missed. Therefore, a mixture of phased and unphased SNV regardless of the zygosity were segmented with the same procedure to capture potential ROH regions. For singleton or duo cases, joint calling can be performed with proxy parental data to generate a joint VCF file, enabling combined CNV and BAF analyses. CNV classification and built-in filter module – FindCNV Based on the benchmark performance on 1000 Genomes Project data, and internal GS data analysis, we included a built-in CNV calling module that will classify the segments with log2 ratio into homozygous deletion [log2 ratio value < −4], heterozygous deletion [log2(0.45)-log2(0.55)] and duplication [log2(1.35)-log2(1.65)], normal [log2(0.9)-log2(1.1)], triplication [log2(1.8)-log2(2.2)] and multiple gains [log2 above 1.175]. Segments with log2 ratio outside the above ranges are classified as unknown copy-number for further assessment. In male cases, short reads in the pseudoautosomal regions (PAR1 and PAR2) on chromosome Y are mapped by default to the PAR reference region on chromosome X. This results in an increased log2 ratio at each PAR, which should not be interpreted as a biological duplication. These technical artifacts do not affect female cases. The read depth visualization will only work on the unique sequence region of chrY for male cases. Potential de novo CNVs in autosomes (chr1-chr22) will be annotated if the aberrant copy number segment is only present in proband genome by comparing genomic segments between proband and parental segments using ‘GenomicRanges::findOverlaps’ function. Homozygous duplications can be distinguished from in cis triplications during analysis based on the expected duplication log2 ratio inherited from each parent which may be supported by the presence of ROH overlapping the CNV [7]. In contrast, homozygous deletions will be characterized by the absence of log2 ratio in proband and log2 ratio consistent with a heterozygous deletion from each parent. Following CNV classification, the built-in filter will report CNV measuring > 10 kb that do not overlap with annotations in the gap track from UCSC genome browser. Coordinates from SDs were obtained from the UCSC Table Browser and merged if they map less than 1 kb apart. If 98% of a CNV call overlaps with SDs, the call will be annotated and can be filtered out during analysis to decrease false-positive calls. Phased B-allele frequency plots Allelic imbalance supportive analysis is performed based on unphased and phased BAF. In trio sequencing, biallelic SNVs with > 7 read depth in all individuals are genetically phased based on Mendelian expectation and classified into five categories, phased to maternal (P1) allele, phased to paternal (P2) allele, non-phased homozygous allele, variant deviates from Mendelian expectation, and non-informative allele. All classified variants are then segmented and visualized with log2 ratio in VizCNV. The segmentation is performed independently for phased B-allele on P1 and P2 site and potential ROH block on P1 and P2 site plus non-phased homozygous site with the CBS algorithm [16]. For ROH inference, VizCNV adopts an established approach based on BafCalculator (https://github.com/BCM-Lupskilab/BafCalculator) [6], which was originally developed for ES data and later optimized and validated on genome sequencing data. The updated implementation demonstrates a positive predictive value of 90% and a true-positive rate of 72% using a segmentation mean parameter of 0.47 [8]. For singletons or duos, variant classification is omitted, and only ROH blocks with allelic origin are visualized together with log2 ratio data. User interface of VizCNV An interactive user interface is built to guide the user to upload the required files to run VizCNV. The input files come from a typical GS pipeline (Additional file 1: Fig. S1a). The genomic analysis can be divided into three major views, genome-wide view, chromosomal view, and table view that allows for direct inspection of variable ranges (Additional file 1: Fig. S1b). Runtime and computational resources The computational performance of the VizCNV workflow was evaluated to assess usability and scalability. The most time-consuming step involves generating binned read depth information from sequence alignment files (BAM or CRAM). This process can be efficiently performed using existing tools such as Mosdepth [15]. Another computationally intensive step is joint variant calling, which can be implemented using multiple established approaches, e.g., GATK. The runtime and memory usage for each major step of the VizCNV workflow was included (Additional file 2: Table S1), providing practical guidance for scaling analyses across datasets of varying size. Cohort description To evaluate the performance of the duplication-detection pipeline, we included a cohort of probands with known structural variants that had been molecularly resolved in prior studies. This cohort consisted of 71 individuals with a confirmed diagnosis of MECP2 Duplication Syndrome (MRXSL; MIM#300260). Their structural rearrangements were characterized using a combination of high-resolution genomic assays, including aCGH, optical genome mapping, and both sr-GS and lr-GS. Detailed methodological descriptions and cohort information are reported elsewhere [20, 21]. In addition, this study includes a cohort of 39 families recruited between 2014 and 2024. Texas Children’s Hospital was the primary site of enrollment but there are families from Chile, Argentina, Mexico, Peru, Canada, the United Kingdom, and California [22–25]. These participants remained 'unsolved' for a potentially contributing pathogenic variant after previous ES. Each subject in this cohort had clinical and laboratory evidence consistent with underlying PIDD. Written informed consent was obtained from all participants. sr-GS was performed on probands and parents to further investigate the underlying molecular cause. Short read genome sequencing Sr-GS was conducted at the Human Genome Sequencing Center (HGSC) at Baylor College of Medicine (BCM) as described in published studies [20, 21]. The PIDD cohort were enrolled as part of the Center for Mendelian Genomics (CMG) and Genomics Research to Elucidate the Genetics of Rare Diseases (GREGoR) initiative (https://gregorconsortium.org/). Briefly, following quality control, sequencing libraries were prepared using KAPA Hyper reagents and sequenced to produce 150 bp paired-end reads. The samples were organized into multiplexed pools to achieve an average coverage of 30x, utilizing the Illumina NovaSeq 6000 system. This approach resulted in an average sequencing yield of 108 Gb per sample, with 96% of the genome covered at a depth of 20 × or greater. To ensure sample identify and integrity the Fluidigm SNPtrace™ method for rapidly genotyping 96 SNP sites was employed to verify gender prior to sequencing and to detect contamination. Using this assay sample identity was verified using the Error Rate In Sequencing software developed at the HGSC. SV calls with Parliament2 The latest version of Parliament2 (0.1.11) [26] was run inside a docker container using the following command ‘docker run dnanexus/parliament2:latest –bam input.bam –bai input.bam.bai -r ref.fa –fai ref.fa.fai –breakdancer –breakseq –manta –lumpy –delly_deletion –delly_insertion –delly_inversion –genotype’. The command will run five precompiled SV callers, Breakdancer, Breakseq, Lumpy and Manta with tuned parameters [26]. CNV calls with CNVpytor CNVpytor (v1.3.2) [27] was applied to the 1000 Genomes Project data using a 100 bp bin size under default parameters. CNV calls located within 1 kb of reference genome gaps, smaller than 2 kb in length, containing > 50% ‘N’ bases in the reference sequence, or supported by less than half uniquely mapped reads were excluded. The remaining high-confidence CNV calls were then compared with consensus calls extracted from the original study. Results Overview of the joint BAF and CNV analysis with VizCNV To prioritize potential homozygous and de novo CNVs from trio sr-GS data, we developed an algorithm to call these variants based on joint BAF and CNV analysis of trio WGS data. There are multiple parameters affecting the sensitivity and specificity of the procedure, while VizCNV allows for customization in all these parameters, we sought to establish a baseline that optimized the precision and recall and can be ‘plugged-in’ and implemented. The BAF (Fig. 1a) and read depth analysis (Fig. 1b) were tuned separately before the joint analysis (Fig. 1c). For read depth analysis, the per window read depth value were normalized, segmented, and classified into heterozygous deletion, heterozygous duplication, homozygous deletion, triplication, multiple gain, normal, or unknown, based on log2 ratio. The performance of CNV classifications were then benchmarked with merged CNV calls from the 1000 Genomes Project [28] to optimize parameters like mappability, range of the log2 ratio, and size of CNV. Predicted inheritance of the CNVs was then classified by comparing the region between proband and parents (Fig. 1a). For the phased BAF analysis, we tuned the ROH and allelic imbalance calling on one of the 1000 Genomes Project trios (HG00405, HG00404 and HG00403). The inheritance was then reclassified with the combined phased BAF data and CNV data.Fig. 1Overview of Integrating BAF and CNV Analysis Using VizCNV. a Demonstrates the methodology for trio-based joint calling and the classification of B-alleles. b Outlines the approach for analyzing log2 ratios and categorizing CNVs. c Illustrates the combined analysis and visualization process using phased BAF and log2 ratio data for de novo duplication and deletion, homozygous duplication and deletion (left to right). For a de novo duplication occurring on the maternal allele, we would expect a log2 ratio of log2(3/2) in the proband's read depth plot, accompanied by phased BAF values of approximately 0.66 for SNVs phased to the maternal allele and 0.33 for those phased to the paternal allele. For a de novo deletion on the maternal allele, we would anticipate a log2 ratio of log2(1/2) in the proband's read depth plot, with the paternal phased BAF near 1, resembling a ROH similar in size to the deletion, or an allelic imbalance pattern if the deletion is mosaic. In the case of a homozygous duplication, we would expect a log2 ratio of log2(3/2) in both parents and log2(4/2) in the proband, with the duplication overlapping a larger ROH region. For a homozygous deletion, the expected log2 ratio would be log2(1/2) in both parents, with no signal in the proband, and the deletion would overlap a larger ROH region Overview of Integrating BAF and CNV Analysis Using VizCNV. a Demonstrates the methodology for trio-based joint calling and the classification of B-alleles. b Outlines the approach for analyzing log2 ratios and categorizing CNVs. c Illustrates the combined analysis and visualization process using phased BAF and log2 ratio data for de novo duplication and deletion, homozygous duplication and deletion (left to right). For a de novo duplication occurring on the maternal allele, we would expect a log2 ratio of log2(3/2) in the proband's read depth plot, accompanied by phased BAF values of approximately 0.66 for SNVs phased to the maternal allele and 0.33 for those phased to the paternal allele. For a de novo deletion on the maternal allele, we would anticipate a log2 ratio of log2(1/2) in the proband's read depth plot, with the paternal phased BAF near 1, resembling a ROH similar in size to the deletion, or an allelic imbalance pattern if the deletion is mosaic. In the case of a homozygous duplication, we would expect a log2 ratio of log2(3/2) in both parents and log2(4/2) in the proband, with the duplication overlapping a larger ROH region. For a homozygous deletion, the expected log2 ratio would be log2(1/2) in both parents, with no signal in the proband, and the deletion would overlap a larger ROH region Benchmark read depth analysis with 1000 Genomes Project CNV calls We selected 10 unrelated individuals from the 1000 Genomes Project (Additional file 2: Table S2) with average depth of coverage of 31.5 × mapped to GRCh38, for which raw alignment files were accessible through the data portal (https://www.internationalgenome.org/data, Additional file 2: Table S3). CNV calls for these genomes were extracted from the initial study [29]. Notably, the original dataset exhibited a bias, with a significantly higher number of deletions ( n = 8,584) compared to duplications (n = 54). An optimized performance was achieved when applying a mappability filter of q30 (Additional file 1: Fig. S2a&b). Comparison across CNV types and size ranges revealed distinct performance patterns (Additional file 1: Fig. S2c). For heterozygous deletions, CNVpytor showed higher precision for smaller CNVs (2–10 kb; 55% vs. 18%), whereas VizCNV demonstrated superior precision for larger deletions (> 10 kb; 76% vs. 41%). Recall for deletions was comparable or slightly higher in CNVpytor (62% vs. 48% for 2–10 kb; 87% vs. 82% for > 10 kb). For heterozygous duplications, both tools showed low precision for smaller events (< 2%), but VizCNV outperformed CNVpytor for larger duplications (> 10 kb; 13% vs. 2%). Recall rates for duplications were similar between the two tools (48% vs. 37% for 2–10 kb; 84% vs. 84% for > 10 kb). Overall, VizCNV demonstrated improved precision for larger CNVs, particularly deletions, while CNVpytor retained higher sensitivity across most size ranges. Phased B-allele analysis with trio using GATK joint genotyping In the joint genotyping analysis, we identified 5,019,224 biallelic single nucleotide variants (SNVs) across the autosomes (chromosomes 1–22, mapped to GRCh38) within the HG00405 family. By comparing genotypes between the proband and the parents, we classified 34% (n = 1,750,252) of the SNVs as informative, attributable to one of the parents. About 54% (n = 2,725,145) of the SNVs are classified as homozygous for alternative (1/1, n = 1,518,744) or reference (0/0, n = 1,206,401) alleles (i.e. nonphased homozygous which indicate potential ROH signals) (Fig. 2a, b). Approximately 0.7% (n = 36,123) of the SNVs deviate from Mendelian inheritance expectations. This number is significantly higher than the expected for biological de novo mutation rate from sr-GS [30, 31] which is around 60–80 SNVs per generation. Therefore, we consider most of these deviations to be genotype errors/false discovery from the GATK joint genotype caller (Fig. 2b). The false discovery rate (FDR) in regions that do not overlap with any segmental duplications (SDs) or RepeatMasker elements is lower and estimated at 0.2%, compared to the overall rate of 0.7%. Notably, and as expected [19], regions annotated as SDs or some regions annotated with RepeatMasker exhibited a significant high increase in FDR (odd ratio > 10, p-value < 10–25) (Fig. 2c, Additional file 2: Table S4).Fig. 2Phased BAF Analysis in a High-Coverage 1000 Genomes Project Trio. a Classification principles for five categories: P1 (phased to parent 1), P2 (phased to parent 2), Mendelian deviation, nonphased homozygous (hmz) site and non-informative, explaining the assignment process. b The average distribution of these categories across chromosomes 1 to 22 in the trio (HG00405, child; HG00404, mother; HG00403, father), with error bars representing the 95% confidence interval. c The odds ratio for the false discovery rate in regions masked for repeats and segmental duplications compared to unique genomic regions Phased BAF Analysis in a High-Coverage 1000 Genomes Project Trio. a Classification principles for five categories: P1 (phased to parent 1), P2 (phased to parent 2), Mendelian deviation, nonphased homozygous (hmz) site and non-informative, explaining the assignment process. b The average distribution of these categories across chromosomes 1 to 22 in the trio (HG00405, child; HG00404, mother; HG00403, father), with error bars representing the 95% confidence interval. c The odds ratio for the false discovery rate in regions masked for repeats and segmental duplications compared to unique genomic regions Joint BAF and log2 ratio analysis on HG00405 family From the original benchmark set for HG00405 [28], we curated four heterozygous duplications and 18 heterozygous deletions as high-confidence CNVs for evaluation. Using VizCNV’s read depth analysis, we achieved 100% recall for duplications (4/4) and 83.3% recall for deletions (15/18) (Additional file 1: Fig. S3a). Independent read depth analysis of HG00405 data called 23 heterozygous duplications and 33 heterozygous deletions with length > 10 kb in assayable regions of the genome. Manual inspection with VizCNV and Integrative Genomics Viewer (IGV) (Additional file 1: Fig. S3b) identified 53.6% (30/56) CNVs that map to uncharacterized problematic regions (see the definition of problematic regions in the Methods section), likely representing false positive calls therefore excluded from further analysis. In total, 26 CNVs were retained for read depth and BAF analysis. The most common false positive cases constitute CNV in regions containing two distinct SDs within a 1 kb – 4 kb interval of unique sequence. Out of 26, 10 CNVs, seven heterozygous duplications and three heterozygous deletions, have at least two phased SNVs from either parent, enabling joint log2 ratio and BAF analysis (Additional file 1: Fig. S3b, Additional file 2: Table S5). Of the three heterozygous deletions, the origin of two can be inferred from BAF (one maternal, one paternal) which showed consistency with read depth analysis. Among the seven heterozygous duplications, the origin could be inferred for six (two maternal, four paternal) using read depth analysis only. The one duplication that failed the read depth classification had a log2 ratio of 0.73 in the mother, which exceeded our set threshold for duplications [log2(1.35)-log2(1.65)], resulting in its classification as undetermined (UND) rather than as a maternally inherited duplication. The phased BAF plot independently confirmed five of the seven duplications, with average BAF ranges of 0.64–0.72 (expected 0.66) on the duplicated alleles and 0.28–0.39 (expected 0.33) on the non-duplicated alleles. One of the BAF non-determined duplication, approximately 1 Mb in size, had a log2 ratio of 0.58 in both the proband and the father, suggesting a paternally inherited duplication. The average phased BAF for the duplicated alleles was 0.48 which is more similar to copy-number neutral alleles (expected 0.5), while the non-duplicated alleles had a phased BAF of 0.33 (Additional file 2: Table S6). Intriguingly, the BAF value of paternal SNPs was consistent with paternal inherited duplication at the proximal duplication region and then oscillates between 0.33 and 0.66 at the distal duplication region. The oscillation pattern suggests the presence of tri-allelic haplotypes at the distal duplicated region (Additional file 1: Fig. S4). That could have resulted from an inherited interchromosomal duplication or from recombination within the duplicated region. In all, combined read depth and BAF analysis confirmed 70% of the curated CNV calls. Detection of CNVs in the Xq28 duplication cohort We specifically benchmarked VizCNV against split-read and discordant read-pair-based SV callers run by Parliament2, on duplications from a cohort with validated CNVs spanning MECP2 at chromosome X varying in size from 64.6 kb to 15.7 Mb (n = 71). The genomic structures were classified as tandem duplications (n = 36) terminal duplications (n = 17), and complex genomic rearrangement (CGR; n = 18) [20, 21]. Read depth analysis facilitated by VizCNV detected all 71 duplications (100%) whereas split-read and discordant read-pair-based SV callers detected 27 out of 71 (38.0%). (Additional file 2: Table S7). The low detection rate of SV callers was dependent on the genomic structure. Of the 35 terminal duplication and complex genomic rearrangements (CGRs), SV callers detected only 5 (14.3%) whereas 22 out of 36 (61.1%) tandem duplications were correctly detected. This result suggests that complex genomic structures contribute to higher false-negative rates for pathogenic CNVs if the analysis is only based on SV callers that focus on split-reads and discordant read-pair information. Among the 13 de novo duplication events, one occurred in a female proband, which allowed BAF and read depth integrated CNV analysis (Additional file 1: Fig. S5). CNV analysis on a pilot primary immune deficiency disease cohort Through chromosomal abnormality analysis with VizCNV, we identified a trisomy 21 in proband BH10262-1 which was initially observed by clinical Chromosomal Microarray Analysis (CMA) but missed from ES analysis, thus, acting as positive control for the algorithm for chromosomal investigation using genome-wide median option for normalization. Detailed BAF and log2 ratio inspection supports a meiosis I non-disjunction event (Additional file 1: Fig. S6). VizCNV reported on average 160 CNVs per family (Fig. 3b) that did not overlap with the SD filter criteria and revealed seven potential de novo CNVs for the whole PIDD cohort (n = 39 trios) (Fig. 3a). Further variant prioritization focusing on an internally curated PIDD gene list [32–34] narrowed this number to 48 inherited rare CNVs (i.e. frequency < 0.05 in gnomAD SV 2.1 database) (Fig. 3a, Additional file 1: Fig. S7). Classification as likely true positive CNVs was based on IGV inspection (Fig. 3, Additional file 2: Table S8) for read depth from proband and parent samples data and breakpoint junction analysis of soft-clipped reads when they are present and BAF distribution. Of the 48 true positive CNVs detected in 39 probands, 27 heterozygous duplications and nine heterozygous deletions, each had at least two phased SNVs from a parent, creating a reliable list for evaluating the performance of log2 ratio and BAF analysis. Among the nine identified deletions, six exhibited log2 ratio within the expected range for heterozygous deletion (HET_DEL) [log2(0.45)-log2(0.55)], classifying them as such, while three deletions fell outside and were classified as UND. All nine deletions can be confidently classified as heterozygous deletions using BAF data, indicating that incorporating BAF can enhance classification beyond relying solely on log2 ratios. Of the 27 identified heterozygous duplications, 20 showed the presence of both paternal and maternal SNVs, while seven exhibited SNVs indicating only one origin (either paternal or maternal) which were not accessed due to insufficient evidence. Among the 20 duplications with informative phased SNPs for both parents, 11 had BAF ranging from 0.62 to 0.70 or 0.29 to 0.37 which are consistent with the expected BAF for duplication. Of these 11, 10 duplications were consistent with the inheritance inferred from read depth analysis (An example shown in Additional file 1: Fig. S8 and Additional file 2: Table S9), with one exception identified in case BH9703-1. This duplication had an average BAF of 0.68 (nine maternal origin SNVs) and 0.33 (21 paternal origin SNVs) (Additional file 2: Table S10). However, the log2 ratio and IGV junction analysis indicated that this duplication was paternally inherited (Additional file 1: Fig. S9). This inconsistency between inheritance predicted by BAF and read depth supports an interchromosomal origin for this inherited duplication or a recombination within the duplication similarly to one of the duplications observed in HG00405. In all, combining read-depth and BAF analysis validated 9/9 deletions and 11/20 duplications.Fig. 3Systematic Workflow for CNV Analysis in PIDD Cohort with Stepwise Filtering and Manual Inspection. a Detailed workflow for the genomic analysis in PIDD cohort consisting of 39 trios. Targeted CNV analysis identified 299 candidate CNVs, which were filtered by removing polymorphic CNVs present in more than five families, resulting in one potential contributing CNV (DOCK8, MIM#243700). De novo analysis identified seven candidate CNVs, which were further inspected using VizCNV and IGV, and further narrowed down to two, with one potential clinically relevant CNV (CSMD1). b Average count of CNV candidates per family progressing through each stage of the selection process Systematic Workflow for CNV Analysis in PIDD Cohort with Stepwise Filtering and Manual Inspection. a Detailed workflow for the genomic analysis in PIDD cohort consisting of 39 trios. Targeted CNV analysis identified 299 candidate CNVs, which were filtered by removing polymorphic CNVs present in more than five families, resulting in one potential contributing CNV (DOCK8, MIM#243700). De novo analysis identified seven candidate CNVs, which were further inspected using VizCNV and IGV, and further narrowed down to two, with one potential clinically relevant CNV (CSMD1). b Average count of CNV candidates per family progressing through each stage of the selection process Genotype and phenotype analysis uncovered in trans pathogenic CNV and SNV in a proband with PIDD PIDD-gene panel target CNV analysis uncovered a paternally inherited 12.8 kb exonic deletion including exon 5 of DOCK8 (Fig. 4). This deletion is present in both proband (BH9040-1) and father (BH9040-3). The deletion was not identified in the initial ES analysis, highlighting the critical advantage of sr-GS in detecting single exon CNVs. Both breakpoints of the deletion map to Alu elements. Microhomology consisting of 24 bp at the junction support an Alu-Alu mediated rearrangement as the mutational mechanism. Of note, the AluAluCNVpredictor (https://alualucnvpredictor.hgsc.bcm.edu/) [35] score for DOCK8 is 0.573; multiple studies have reported different exon deletion alleles for this gene in association with Hyper-IgE syndrome [36–39]. Rare SNV analysis on ES data identified a maternally inherited and ultrarare missense variant (NM_203447.4:c.5599 T > G, p.(Phe1867Val) in DOCK8. The SNV is present only in this family (BH9040-1 and BH9040-2) in the BCM GREGoR genomic variant database and absent in control databases (gnomAD v4.0 https://gnomad.broadinstitute.org/, Mexico City Prospective Study https://rgc-mcps.regeneron.com/home [40], and All of Us https://databrowser.researchallofus.org/snvsindels/). Moreover, the amino acid position exhibits significant evolutionary conservation (PhyloP100way score 7.781). The clinical presentation of the case included elevated IgE levels, IgM deficiency, recurrent infections with herpes viruses and Staphylococcus spp, and T cell deficiency, phenotypically consistent with autosomal recessive Hyper-IgE syndrome (MIM#243700).Fig. 4Compound heterozygous variants on DOCK8. a On top shows the karyogram of chr9, below shows the screenshot of VizCNV chromosomal view including read depth and B-allele frequency plots showing a paternally inherited heterozygous deletion of exon 5 in DOCK8. The read depth graph (log2 ratio plot) shows the heterozygous deletion in the proband and father. Blue: proband, Yellow: mother, Turquoise: father. The BAF plot correlates with the deletion, showing a lack of heterozygous SNPs and maternally phased homozygous B alleles near 1, aligning with the paternal deletion. The OMIM track shows that this variation locates in DOCK8 gene. The bottom track shows that repeat elements distribution across the region. The purple circles highlight the Alu elements overlapping with breakpoints. b On the top shows the transcript structure of DOCK8, noting a maternally inherited missense mutation (NM_203447.4:c.5599 T > G, NP_982272.2:p.F1867V) in exon 44 marked. A detailed view of the deletion’s genomic architecture is provided, alongside the breakpoint junction sequence alignment, which suggests an Alu-Alu mediated deletion. c Segregation of the variant within the family (BH9040) Compound heterozygous variants on DOCK8. a On top shows the karyogram of chr9, below shows the screenshot of VizCNV chromosomal view including read depth and B-allele frequency plots showing a paternally inherited heterozygous deletion of exon 5 in DOCK8. The read depth graph (log2 ratio plot) shows the heterozygous deletion in the proband and father. Blue: proband, Yellow: mother, Turquoise: father. The BAF plot correlates with the deletion, showing a lack of heterozygous SNPs and maternally phased homozygous B alleles near 1, aligning with the paternal deletion. The OMIM track shows that this variation locates in DOCK8 gene. The bottom track shows that repeat elements distribution across the region. The purple circles highlight the Alu elements overlapping with breakpoints. b On the top shows the transcript structure of DOCK8, noting a maternally inherited missense mutation (NM_203447.4:c.5599 T > G, NP_982272.2:p.F1867V) in exon 44 marked. A detailed view of the deletion’s genomic architecture is provided, alongside the breakpoint junction sequence alignment, which suggests an Alu-Alu mediated deletion. c Segregation of the variant within the family (BH9040) For the seven de novo CNV calls from 39 families, subsequent verification using IGV and VizCNV joint BAF and log2 ratio inspection confirmed three high-confidence de novo CNVs – two deletions and one duplication. Notably, two de novo CNVs,one deletion and one duplication, were identified from two different families (BH8461 and BH14227), both occurring within the same genomic regions (chr11:18944000–18964000, ~ 20 kb in size). Both CNVs are flanked by LINE elements, hinting at a potential recurrent CNV mediated by LINE-LINE recombination. This recurrent CNV involves gene MRGPRX1, which encodes for Mas-related G-protein coupling receptor. In addition to the two de novo CNVs, eighteen inherited CNVs were identified at the same locus. One de novo deletion (~ 59 kb) was identified within intron 3 (NM_033225.6) of the gene CSMD1 in BH9703-1. The deletion does not remove any exons of CSMD1 but rather removes a non-coding element ENSG00000286934 which encodes an antisense RNA to CSMD1 (Additional file 1: Fig. S10). Inherited CNVs with inconsistent origin in phased BAF analysis reveal clues about formation mechanisms Consistent parental origin based on log2 ratio and for phased BAF of inherited CNV segments were observed in five out of six duplications in HG00405 family and 10 out of 11 duplications in the PIDD cohort. In those 15 out of 17 alleles (88%) the distribution of the phased BAF supports intrachromosomal duplications, i.e. the two alleles are copied from the same ancestral chromosome and therefore present duplicated BAF from the same parent who carries log2 ratio consistent with a duplication (Fig. 5a). For two out of 17 alleles (12%) distinct patterns emerge, for which we propose different models to explain (Fig. 5).Fig. 5Proposed models explaining inconsistencies between read depth and BAF Inheritance Analysis. The top panel shows the pedigree with duplication segregation information inferred from read depth and IGV analysis. Below the pedigree are the proposed models explaining the observed BAF patterns. The two lines and rectangular segments represent two distinct haplotypes in the proband: one paternally inherited (P) and one maternally inherited (M). The green rectangle indicates the tandem duplication segment. Tick positions mark informative SNPs, with colors indicating phasing information: maternal phase (orange), paternal phase (teal), and unphased (black). The bottom section shows the phased B-allele frequency plot. a Single-step model with intrachromosomal duplication; read depth analysis suggests maternally inherited duplication with expected phased BAF of maternal (0.66) and paternal (0.33) SNPs. b Multi-step model with intrachromosomal duplication and phased BAF swap, indicating a mixture of haplotypes in the duplication region potentially due to meiotic recombination. c Single-step model; read depth analysis suggests paternally inherited duplication but inconsistent phased BAF of maternal (0.66) and paternal (0.33) SNPs, suggesting potential interchromosomal duplication Proposed models explaining inconsistencies between read depth and BAF Inheritance Analysis. The top panel shows the pedigree with duplication segregation information inferred from read depth and IGV analysis. Below the pedigree are the proposed models explaining the observed BAF patterns. The two lines and rectangular segments represent two distinct haplotypes in the proband: one paternally inherited (P) and one maternally inherited (M). The green rectangle indicates the tandem duplication segment. Tick positions mark informative SNPs, with colors indicating phasing information: maternal phase (orange), paternal phase (teal), and unphased (black). The bottom section shows the phased B-allele frequency plot. a Single-step model with intrachromosomal duplication; read depth analysis suggests maternally inherited duplication with expected phased BAF of maternal (0.66) and paternal (0.33) SNPs. b Multi-step model with intrachromosomal duplication and phased BAF swap, indicating a mixture of haplotypes in the duplication region potentially due to meiotic recombination. c Single-step model; read depth analysis suggests paternally inherited duplication but inconsistent phased BAF of maternal (0.66) and paternal (0.33) SNPs, suggesting potential interchromosomal duplication For the ~ 1 Mb duplication observed in HG00405, the presence of variable-size BAF blocks with switching of parental origin suggests a multi-step model involving an ancestral intrachromosomal duplication followed by meiotic recombination, which disrupted the linkage of the haplotype blocks (Fig. 5b). In contrast, for the second duplication (BH9703-1), no haplotype switching was identified within the region, but the duplicated BAF (0.66) is noninformative regarding origin, whereas the BAF markers supposedly not duplicated (0.33) show inconsistent origin compared to the parental log2 ratio (Fig. 5c and Additional file 1: Fig. S9a). This pattern can be explained in two ways, 1) it is a two-step process similar to the duplication in HG00405, where the first step is the intrachromosomal duplication followed by a meiotic recombination that produces a similar haplotype to non-affected allele, or 2) it is a single-step process (Fig. 5c) for which the duplicated segments might have originated from two homologous chromosomes, i.e. interchromosomal duplications [11, 12, 41]. This suggests that intrachromosomal duplications might occur more frequently or they have higher stability, influencing their greater prevalence when compared to interchromosomal duplications. Discussion Analysis of computationally identified CNVs is a critical step to support variant interpretation and investigation of its functional consequences. Examples include gene structure alterations such as formation of fusion genes and splicing effects or mRNA expression dysregulation [11, 12]. Moreover, it reveals molecular features of SV mutagenesis mechanisms [42], and organismal hypermutation processes [5, 43]. Currently, there are thousands of rare disease families with sr-GS data available for analysis in restricted public databases for which an open and free resource such as VizCNV can be applied for further SV investigation. Examples include the GREGoR consortium which comprises 10,683 individuals-76% with sr-GS data (R04, https://gregorconsortium.org/) [44] and Gabriella Miller Kids First Program (https://kidsfirstdrc.org/) which has enrolled 38,180 participants, 25,100 of whom have Illumina BAM or CRAM data available (https://kidsfirstdrc.org/ as of December 2025). As the rare disease projects expand, sequencing availability to more probands and families, CNV analysis pipelines are essential to reduce false positive and false negative variant calls and to improve the detection of CGRs. Open-source tools have been developed to visualize CNVs and SVs greatly reducing false positive calls [27, 45]; However, they often lack BAF phasing, interactive features, or do not enable trio analysis. VizCNV supports an effective first-pass analysis, enabling the identification and prioritization of high-impact CNVs, such as de novo and homozygous variants. Integration of independent SV calls supports the validation process by incorporating breakpoint junction information. As exemplified by the analysis of pathogenic duplications at Xq28, CNVs with breakpoints mapping to genomic repeats are promptly detected by read depth analysis although mostly missed by SV callers based on breakpoint junction detection derived from split-read or discordant read-pair signals. A CNV/SV complementary approach as performed here ensures the detection of genomic rearrangements formed by nonallelic homologous recombination (NAHR) or break-induced replication (BIR) mechanisms both of which causative of several genomic disorders [1]. Similar challenges are posed by CGRs, for which CNV or SV analysis alone may be insufficient for accurate identification. Furthermore, chromosomal or genome-wide copy-number analysis enables the detection of both large-scale aneuploidies and smaller pathogenic CNVs that may be missed by ES. For example, this approach identified a case of trisomy 21 that was initially undetected by ES analysis, as well as a submicroscopic single-exonic CNV in a proband with Hyper IgE syndrome. Compared with existing visualization tools, VizCNV achieves comparable recall with improved precision for CNVs larger than 10 kb. Tools such as Samplot [45] depend on external CNV callers which may miss complex structural variants at broader scales. Importantly, VizCNV provides an interactive visualization framework that supports both genome-wide overview and allows for detailed locus-level inspection, while integrates parent-of-origin information relevant to support CNV calls and for mechanistic studies (Additional file 3: Table S11). Allelic imbalance can be introduced by multiple mutational events, i.e., non-segregation during meiosis, unbalanced translocations, intrachromosomal and interchromosomal rearrangements, etc. [21, 46]. Here, we show that integrated analysis of phased and unphased BAF, SVs and CNVs enables identification of pathogenic variants but also provides mechanistic information. The annotation of phased SNVs with log2 segments facilitates inference of inheritance patterns and provides mechanistic insights into the formation of underlying genomic rearrangement. For example, the occurrence of recombination or template switching generating a specific CNV can be revealed by the presence of intrachromosomal and interchromosomal genomic segments [1]. In fact, the frequency of interchromosomal and intrachromosomal events in the de novo formation of CNVs is mostly unknown. This information is clinically important because interchromosomal copy-number gains can concurrently generate large ROH regions, which increases the risk for recessive diseases and uniparental isodisomy [47]. Moreover, interchromosomal events in autosomes can be associated with CGRs as exemplified by pathogenic inverted triplications and interspersed duplications [11, 42]. Interestingly, for genomic rearrangements generated by microhomology-mediated break induced replication (MMBIR) and BIR in diseases such as Potocki-Lupski syndrome (PTLS; MIM #610883) and Lubs-type X-linked syndromic intellectual developmental disorder (MRXSL; MIM#300260), there is a notable preference for intrachromosomal duplications [41, 48]. In fact, only 10.5% (two out of 19 cases) of pathogenic CGRs at Xq28 involve interchromosomal gains [21], with majority being intrachromosomally generated, likely during paternal germline mitosis. The frequency of interchromosomal gains in chromosome Xq28 aligns closely with the proportion of genome-wide interchromosomal duplications (12%) observed in the PIDD cohort. Further BAF analyses of CNVs are necessary to establish the prevalence of those distinct events genome-wide and their implications to genomic rearrangement formation. In all, the synergistic approach combining phased BAF and CNV greatly extends the utility of sequencing data beyond the identification of mutation and provides mechanistic clues including concurrence of recombination within CNVs, implications for CGR formation and origin of denovo CNVs. VizCNV has been optimized for sr-GS data. One limitation of the tool is its reliance on genetic phasing based on trio genotypes to infer parental origin. The accuracy of such inference is constrained by genotype quality, e.g., increased genotype error in repetitive or low-complexity regions, and by the density of informative heterozygous SNVs available for phasing. While statistical phasing, e.g., population reference panels or physical phasing, e.g., long read or linked-read can extend haplotype blocks beyond trio-informative sites [4, 49, 50], they are not yet incorporated into the current version of VizCNV. Analysis of multiallelic sites without parental information have the potential to distinguish complex haplotype origins and segregations, but more advanced analytical tools and study design are needed [19]. Incorporating these sites could enhance our ability to infer the structure of complex genomic variations and their implications in genetic disorders. We anticipate that future integration of long-read data and multiallelic variant information will enhance VizCNV’s ability to resolve complex haplotype structures, and to inspect allele-specific events involving recombination or template switching. Incorporating these features will expand the functions and analytical scope of VizCNV. Conclusions VizCNV provides a streamlined, open-source platform for sensitive detection and visualization of aneuploidies, CNVs, and SVs from sr-GS. By integrating read depth analysis, BAF, ROH and haplotype-aware analyses, the tool supports CNV interpretation, through improved detection of de novo and inherited variants, establishment of parental origin, and mosaicism. Application to an ES-unsolved rare disease cohort demonstrated its clinical utility, including identification of pathogenic compound heterozygous variants underlying Hyper-IgE syndrome. As GS becomes increasingly central to rare disease diagnostics, VizCNV provides a free and flexible platform for enhancing diagnostic yield and supporting mechanistic studies across diverse clinical and research settings. Its ability to unify CNV calling with intuitive visualization supports more accurate variant interpretation, contributing to precision medicine efforts and gene-disease discoveries. Supplementary Information Additional file 1. Supplementary figure S1-S10. Additional file 2. Table S1-S10. Additional file 3. Table S11. Additional file 1. Supplementary figure S1-S10. Additional file 2. Table S1-S10. Additional file 3. Table S11.