These studies have provided new insights into the genetics that contribute to CVD development. GWAS aim to identify the genetic variants that are commonly associated with different diseases and traits by comparing variant frequencies between the cases and controls. More than a hundred hypothesis-free GWAS have been carried out over the past 15 years to identify the genetic loci associated with the CVD risk (Walsh et al., 2023). The finding of association between SNPs on 9p21 and coronary heart disease or myocardial infarction is the most strongly supported finding among different ethnic groups studied (Ndiaye et al., 2011). These studies were able to identify: a locus on chromosome 9p21 which was associated with coronary heart disease in Caucasian populations (McPherson et al., 2007), the SNPs in the genes PHACTR1, SORT1, LDLR and PCSK9 are associated with coronary heart disease and myocardial infarction (Myocardial Infarction Genetics et al., 2009; Aragam et al., 2022), SNPs near lipid regulating genes such as CYP7A1, NPC1L1, SCARB1, PPP1R3B and SLC39A8 lead to CVD risk factors such as higher total cholesterol levels in individuals with European ancestry (Teslovich et al., 2010; Waterworth et al., 2010), the mutations in LDLR, PCSK9 and APOB genes cause familial hypercholesterolemia which is an inherited condition that result in higher cholesterol levels in the blood (Doi et al., 2021; Marmontel et al., 2023). Also, SNPs in genes SH2B3, ATP2B1, CYP17A1, CYP1A2, MTHFR, ZNF652, PLCD3 and FGF5 are associated with high blood pressure (Newton-Cheh et al., 2009; Levy et al., 2009). While GWAS have identified thousands of single nucleotide polymorphisms (SNPs) that are associated with CVD, most of these SNPs (more than 80%) reside in the noncoding region of the genome, therefore their functional significance remains unknown (Tam et al., 2019; Walsh et al., 2023; Ndiaye et al., 2011). While GWAS have been instrumental in identifying common variant associations, they capture only a fraction of the genetic architecture of CVD. This gap has motivated increasing interest in SVs, which can exert larger functional effects and may account for part of this unexplained variance.

Among the various classes of structural variation, copy number variants (CNVs) are the most extensively studied and represent a major contributor to dosage-related cardiac phenotypes. When the deletion, insertion and duplications cause variations in copy numbers compared to a reference genome these are called CNVs (Collins et al., 2020). Inversions and translocations do not cause CNVs in the genome (Collins et al., 2020). Deletions cause the loss of DNA segments from the genome; if they disrupt a coding region or regulatory region this can lead to a loss of function effect of a particular gene or several genes (Escaramís et al., 2015). Mademont-Soler et al. were able to identify and validate a MYBPC3 deletion (entire exon 27 and a deletion spanning from exon 4 to exon 12) in 2 patients and a PLN gene coding region deletion in 2 patients out of 303 unrelated Spanish patients with hypertrophic cardiomyopathy (HCM) diagnosis (Mademont-Soler et al., 2017). Another genome-wide study was able to identify an 8733 bp deletion spanning the exon 4 of the BAG3 gene in seven family members with dilated cardiomyopathy via comparative genomic hybridization (CGH) which was absent in 355 controls (Norton et al., 2011). Duplications result in the doubling of the genome regions, when the duplication occurs in the same region contiguously without any intervening sequence, it is called a tandem duplication (Van Dyke et al., 1983). A study by Lopes et al. was able to find a large deletion in the genes MYBPC3 and PDLIM3 and a large duplication in TNNT2 and LMNA genes in 505 unrelated HCM patients using targeted sequencing methods focusing on the genes related to cardiomyopathy (Lopes et al., 2015). Recent studies have also shown the involvement of CNVs in CVD such as myocardial infarction (Shia et al., 2011), stroke (Matarin et al., 2008), coronary artery disease (Wu et al., 2014) and congenital heart diseases (Glessner et al., 2014). Different CNVs play a significant role in CVD in different populations and their frequency of occurrence may vary from one population to another (Pollex and Hegele, 2007). Overall, increasing evidence suggests that CNVs play an important role in CVD development, but larger population studies with WGS technologies will be required to better understand the clinical significance of these CNVs (Vijay et al., 2018). Beyond CNVs, structural variation also arises from mobile element activity, which generates insertions and complex rearrangements with important implications for CVD-related regulatory disruption.

Insertion is an addition of a new DNA segment into the genome which may either disrupt genes or their regulatory regions, such as enhancers, silencers and topologically associated domains (TADs) (Escaramís et al., 2015). MEs are mostly responsible for inserting DNA fragments into the genome. These elements are repetitive DNA sequences capable of inserting themselves into different locations of the genome and contributing up to 42% of the human genome (Prak and Kazazian, 2001). MEs fall into two main categories (Figure 2): DNA transposons, which follow a ‘cut and paste’ mechanism to move and retrotransposons, which use an intermediate RNA molecule to move and will be then reverse transcribed to DNA before inserting into the genome. Retrotransposons can be further divided into long terminal repeats retrotransposons (LTR transposons) and non-LTR transposons which include long interspersed nuclear elements (LINES) (e.g., L1 element) and short interspersed nuclear elements (SINES) (e.g., Alu elements) (Prak and Kazazian, 2001). As these elements are heritable and dispersed throughout the genome, different loci could contain similar repetitive sequences susceptible to different mispairing and recombination events that lead to deletions or insertions in the genome (Kazazian, 1998; Xing et al., 2009; Kojima et al., 2023).

Studies have also shown the involvement of MEs in CVD development. Kataoka et al. found that exon 1 and 3 in the BMPR2 gene have been deleted due to a recombination between two AluY elements in patients diagnosed with pulmonary arterial hypertension (Kataoka et al., 2013). Another study showed that the insertion of 333bp Alu Ya5 element in the ALMS1 gene causes the Alstrom syndrome in a large Turkish family (Taskesen et al., 2012). Furthermore, the study of Holdt et al. was able to find that the Alu elements present in the ANRIL noncoding RNA, encoded by chromosome 9p21 which has been identified as the strongest genetic factor of atherosclerosis, are important in regulating the atherogenesis processes (Holdt et al., 2013). There are only very few studies that have been carried out to identify how MEs are involved in CVD development, therefore more investigations are needed to identify the effect of MEs on various CVD subtypes. In addition to ME-mediated events, several chromosome-scale rearrangements, including inversions and translocations, also contribute to structural complexity relevant to cardiovascular phenotypes.

When a DNA segment breaks and is reattached to the same chromosome or different chromosome of the genome in an opposite direction it is called an inversion. This type of SV interferes with the chromosome pairing events during meiosis and potentially leads to recombination events (Kirkpatrick and Barton, 2006). With respect to CVD one study found that polymorphic inversion in chromosome 17q21.31 is involved in higher cardiac expression of CRFR1 in heart failure patients (Pilbrow et al., 2016). Furthermore, few studies have also been carried out to identify the inversions that are related to CVD risk factors such as obesity (Gonzalez et al., 2020; Caceres and Gonzalez, 2015) and diabetes (Antonacci et al., 2009). When a DNA segment changes its location interchromosomally or intrachromosomally, it is called a translocation (Escaramís et al., 2015). The translocation can also cause disruption or dysregulation of genes depending on the chromosomal breakpoint. For example, Curran et al., identified a translocation that disrupts the elastin gene on chromosome 7 which is associated with the development of supravalvular aortic stenosis disease (Curran et al., 1993). They were also able to identify a 300bp Alu repeat within the intron 27 of the elastin gene when they were analyzing exon 28, which was the breakpoint of this translocation, suggesting that Alu elements may have been involved in the translocation mechanism (Curran et al., 1993). Generally, all the above variant types generate two breakpoints in the genome and, different methods are being used to identify these breakpoints and characterize the SV type.

Many SVs can exceed the length of the short reads generated by ‘short read’ WGS technologies, thereby increasing the difficulty of detecting SVs and SVs present in the repetitive regions (Escaramís et al., 2015; Mahmoud et al., 2019; Liu et al., 2024). The introduction of ‘long read’ WGS technologies has overcome this read length challenge and has the advantage of detecting complex SVs and SVs in repetitive regions over the other methods (Yang, 2020; Liu et al., 2022). Since short reads are preferred over long read WGS technologies for larger data sets and population data research due to their relatively low cost and high throughput, to facilitate the identification of SVs using short reads four main conceptual strategies have been developed.

These include Read Depth, Read Pair, Split Read and Assembly Methods. The Read Depth (RD) method identifies the unexpected changes in the read count (depth of coverage) in the genomic regions and is able to detect copy number variants but is limited in breakpoint detection (Yang, 2020; Zhao et al., 2013). The read pair (RP) method utilizes the orientation, and the span of paired-end reads to identify the inconsistencies between the reference genome and the read pairs, but it is known to be less effective in repetitive regions (Alkan et al., 2011). The split read (SR) method detects the paired reads where one read mapped completely to the reference genome while the other read fails or is only partially mapped to the reference; the reads that fail to map are used to identify the breakpoint of SVs (Jakubosky et al., 2020). The assembly method (AS) generates contigs by de novo assembling the reads and the contigs are then mapped to the reference genome to identify the SVs (Guan et al., 2016). The assembly method can detect more SVs compared to other methods, but it is limited in use due to its high demand for computational resources (Alkan et al., 2011). Taken together, these strategy-specific strengths and limitations underscore the importance of evaluating SV discovery holistically rather than tool-by-tool.

Across SV discovery workflows, three recurring constraints determine downstream interpretability: (i) repeat content and reference bias, (ii) breakpoint precision, and (iii) cohort scalability. Short-read pipelines (RD/RP/SR/AS) remain cost-efficient and scalable for population studies yet systematically underperform in repeat loci and in detecting novel insertions, leading to ambiguous breakpoints that weaken mechanistic inference and reduce association power across cohorts (Ho et al., 2020; Mahmoud et al., 2019). Long-read and HiFi approaches mitigate these issues with improved breakpoint resolution and enhanced detection in complex or repetitive regions but are still limited in large biobank-scale studies due to cost and throughput (Yang, 2020; Liu et al., 2022). These trade-offs motivate hybrid strategies, such as discovery using short reads followed by targeted long-read or optical validation (Escaramís et al., 2015; Liu et al., 2024). At the caller level, combining multiple evidence types, RD for CNVs, RP for insertions/inversions, SR for precise breakpoints, and AS for complex alleles, consistently outperforms single tools, provided merging uses calibrated confidence thresholds and orthogonal validation matched to SV class/size (Kosugi et al., 2019; Alkan et al., 2011). These methodological choices directly shape clinical translation, influencing gene-level attribution, pathogenicity classification, and diagnostic yield (Pagnamenta et al., 2023; Stranneheim et al., 2021) as well as epidemiological analyses, where breakpoint uncertainty and reference bias affect genotype concordance, and allele frequency estimation (Collins et al., 2020; Sudmant et al., 2015). Building on these conceptual approaches, a diverse ecosystem of SV callers has been developed, each leveraging different combinations of RD, RP, SR, or assembly signals to optimize detection across SV types.

Numerous bioinformatics tools have been developed to detect SVs in the sequencing output data (Table 3). Some tools are specialized in detecting SVs in tumor samples and some are specialized in the input data types such as ‘short read’ and ‘long read’ sequences. Every tool shows different levels of precision and recall rates in calling SVs of different types, different sequencing depths and different SV size ranges (Kosugi et al., 2019; Mahmoud et al., 2019).

RD based tools such as CNVnator (Abyzov et al., 2011) and Canvas (Roller et al., 2016) and RP and SR based methods like Delly (RP, SR) (Rausch et al., 2012), LUMPY (RD, RP, SR) (Layer et al., 2014), Manta (RP, SR) (Chen et al., 2016) are widely being used for identifying SVs in WGS data. In a study to identify de novo variants and novel candidate genes related to congenital heart disease, Manta and Canvas tools were used to call CNVs (Hartill et al., 2024). In addition to the above approach-based tools, tools based on machine learning are also being developed to overcome the problems associated with approach-based methods such as caller-specific tools, problems associated with the sequencing properties and alignment algorithms. Cue (Popic et al., 2023) DeepSV (Cai et al., 2019) and SVision (Lin et al., 2022) are a few examples of SV calling tools, DeepSVFilter (Liu et al., 2021) is a false positive SV filtering tool and NPSV-deep (Linderman et al., 2024) is a SV genotyping tool developed based on deep learning algorithms and these tools can detect, genotype and filter a wide range of SV types in WGS data.

However, there are limitations and drawbacks associated with each tool, for example, tools with high precision could have missed true positives and tools with higher specificity could result in many false positives. Some tools require higher computational power and memory, and some tools may struggle in repetitive genomic regions to detect SVs leading to inaccurate results. Therefore, determining the optimal tools for a specific study remains a challenge both computationally and scientifically. Since each of these tools has its own strengths and weaknesses, there is no gold standard tool that can capture all the types of SVs with higher accuracy (Tattini et al., 2015). Hence, the detection of SVs requires integrated and reproducible strategies, the selection of SV detection tools should rely on the data types, study requirements, complexity of the genome and the type of SVs of interest (Mahmoud et al., 2019; Joe et al., 2024). Also, as many studies have suggested, for more comprehensive SV detection use of a consensus or ensemble approaches with complementary evidence types would be the best practice as it combines the capabilities of various tools to improve overall accuracy (Kosugi et al., 2019; Yang, 2020; Joe et al., 2024). For example, using tools that use RD strategy to detect copy number changes, SR for precise break points and RP for insertions and inversions, could offer more deeper insights into complex SVs rather than using multiple callers that use the same strategy (Kosugi et al., 2019; Tattini et al., 2015). Also, merging these outputs requires careful considerations, such as confidence thresholds (support from more than 2 callers or evidence types) and categorizing these calls into high confidence based on multiple signal support (Liu et al., 2024). For more accurate results, validation strategies should also be orthogonal and specifically matched to SV type and size; small SVs (<1 kb) are best validated by PCR or Sanger sequencing, while medium SVs (1–100 kb) are commonly validated by quantitative PCR, ddPCR, long read sequencing and optical mapping methods and larger complex SVs (>100 kb) are benefit from cytogenetics methods (FISH, karyotyping) or linked-read/Hi-C data (Ho et al., 2020). Furthermore, benchmarking with high confidence and well characterized truth sets (e.g., HGSVC, GIAB) and report precision, recall, breakpoint accuracy and genotype concordance stratified by SV type, size and genomic context (including repeat rich regions, telomeric and centromeric regions as these often possess particular challengers for accurate SV detection) are essential to reveal systematic biases, to assess method performances and to better understand the strengths and limitations (Liu et al., 2024). Pipelines should also include pre-filtering (based on minimum read support, quality thresholds) and post-processing (genotype refinement and annotation using population databases such as gnomAD-SV) to provide a rigorous and reproducible framework for SV discovery. By following these best practices, and documenting all software versions, parameters and reference genome build, can make SV calling workflows more robust and reliable, ensuring reproducibility and transparency. Also, use of new genome assemblies like pangenome and the use of artificial intelligence could enhance the accuracy and reliability of SVs detection, helping to refine methods, improve consistency and meaningful comparison in SV research across different datasets and platforms thereby improving our understanding of the genetic underpinning of CVD which could facilitate improved risk assessment strategies. While caller performance is central to accurate SV discovery, population-scale interpretation requires additional considerations related to genotyping, allele-frequency estimation, harmonization, and statistical modelling.