Long-read reference panel estimates structural variants from existing SNP data and maps associations with complex traits
Background: Studying structural variants using existing SNP data
A study published in Nature Genetics used a reference panel built from long-read genomes to estimate structural variants (SVs) from existing single-nucleotide polymorphism (SNP) data and link them to human complex traits. The team developed ImputeSV and applied it to 456,643 participants of European ancestry in the UK Biobank (UKB). The work provides a basis for analyzing SVs that are difficult to measure directly in large populations using existing data.
SVs generally include insertions, deletions, and inversions of 50 base pairs (bp) or more. SNP arrays can detect only some large SVs. Short-read whole-genome sequencing identifies more SVs, but still faces difficulties in complex regions such as repetitive sequences. Even highly accurate long-read sequencing is costly to apply across an entire large cohort. This study used high-accuracy genomes from a relatively small group as a reference to broaden the scope of large-scale SNP data.
Key findings: Linking 241 high-accuracy genomes to an analysis of 450,000 people
The team used 482 assemblies generated from PacBio HiFi long-read data for 241 people from diverse ancestry groups. From data that distinguish the two genomes inherited from each person's parents, the researchers obtained 171,233 high-quality SVs. They built an imputation reference panel to estimate unmeasured SV genotypes based on patterns of co-occurrence between SNPs and SVs.
Performance varied with the input data and evaluation metric. When array-measured SNPs were first imputed and then used as input, the method achieved an F1 score of 0.92 against Genome in a Bottle (GIAB) reference data and genotype concordance of up to 0.97. F1 evaluates detection precision and recall together. At the population level, the median squared correlation coefficient (r²) between imputed and observed genotypes for common SVs was 0.78. This r² evaluation excluded admixed American and South Asian groups because their sample sizes were insufficient.
In the UKB analysis, the method imputed 54,578 common SVs with a minor allele frequency of at least 1%. A genome-wide association study (GWAS) of 2,624 traits identified 17,335 associations between 4,397 SVs and 656 traits. At 958 loci that met at least one criterion based on lead-variant status, conditional analysis accounting for small variants, or fine-mapping, small genetic variants alone were unlikely to explain the association signals.
In a separate joint analysis of common SVs and small variants across 636 quantitative traits, the two components explained an average of 1.0% and 20.2% of phenotypic variance, respectively. The team estimated that SVs accounted for at least 4.7% of the common genetic variance represented by these two components combined. This figure should be distinguished from a proportion of disease incidence or total heritability.
Significance and outlook: A public research resource for narrowing causal candidates
Trait-associated SVs were enriched in regions regulating gene expression and in RNA-binding protein binding sites, among other features, compared with other SVs. A 60-bp deletion associated with asthma was linked to NRROS expression in tibial artery tissue. A 364-bp deletion within JAZF1 associated with type 2 diabetes was also linked to expression of that gene in several tissues, with the strongest expression association observed in the pancreas. These findings suggest candidate mechanisms for follow-up experiments, but do not establish a functional causal relationship in which the deletions cause disease.
At many loci, linkage disequilibrium—the tendency of nearby variants to be inherited together—makes it difficult to distinguish the effects of SVs from those of small variants. The main analysis focused on people of European ancestry, and estimates for non-European groups were less stable because of smaller samples. The publicly available panel and tools can support reanalysis of existing data, but improvements in clinical diagnosis or genetic risk prediction must be validated separately.
The study design and results can be consulted in the provided Europe PMC full text of the Nature Genetics paper.
Nature Genetics, Published online: 20 May 2026. DOI: 10.1038/s41588-026-02612-z
Summary: This structural genomics study bypasses the resolution limits of traditional short-read sequencing by establishing an advanced reference panel harvested from comprehensive long-read assemblies. The developed web application enables high-fidelity imputation of complex structural variants (SVs) directly from single-nucleotide polymorphism (SNP)-level datasets. Deployed across multi-centric cohorts, this framework unveiled hidden causal linkages between complex traits and genomic rearrangements, introducing a highly scalable, low-cost computational baseline for programmable disease risk stratification and personalized diagnostic modeling.
Research teams with existing SNP array data can use the public ImputeSV reference panel to impute common structural variants and explore their associations with traits before generating new long-read genomes for every participant. For example, if only an SNP signal has been identified in a disease-associated region, analyzing nearby SVs as well can narrow the candidates for follow-up functional experiments. Linking the results to gene expression data may also allow researchers to investigate regulatory processes involving candidate variants. However, imputation performance must first be evaluated for the target population and input data. This study broadens the use of existing genomic data, but did not validate the accuracy of individual risk prediction or the effectiveness of early screening.