01 Nature Communications · 17 September 2026_
European ash pangenome reveals widespread structural variation and genetic basis of low ash dieback susceptibility
- Structural variation
- Population genetics
- Methods
A pangenome built from 50 European ash genomes catalogues structural variants and dispensable genes, then uses it to find 220 SNPs tied to lower susceptibility to ash dieback across more than 1,200 trees.
Summary
European ash is being decimated by ash dieback disease, and a single linear reference genome misses the structural variation that could explain why some trees survive. The authors built a pangenome from 50 geographically diverse Fraxinus excelsior samples, finding 362,965 structural variants including 174 Mb of sequence absent from the linear reference (22% of its size) and 3,412 high-confidence dispensable genes present only in some individuals. They then used the pangenome to reanalyse existing genomic data from over 1,200 trees, finding 220 SNPs with consistent allele-frequency shifts between healthy individuals and those badly damaged by the disease, with the signal holding across separate UK seed sources.
What to take from it
- 50 diverse F. excelsior genomes: 362,965 structural variants, 174 Mb absent from the linear reference (22% of its size)
- 3,412 high-confidence dispensable genes present in only some individuals
- Reanalysis of 1,200+ existing genomic profiles found 220 SNPs with consistent allele-frequency shifts between healthy and badly damaged trees
- The signal held across separate UK seed sources, indicating a shared genetic component to low ash dieback susceptibility
Review
This delivers both a resource and a finding: a pangenome that captures variation a linear reference misses, and an association signal built on it. The association itself is correlative - consistent allele-frequency shifts between healthy and damaged trees, not a validated causal test - and the disease's urgency means results like this could get used in seed selection for restoration plantings before the underlying genes are functionally confirmed. Ash is a long-lived, outcrossing tree, so even a validated marker would take decades to translate into resistant stock at scale. Worth checking before relying on it: whether the 220-SNP signal replicates in independent, non-UK populations, and whether any of the flagged loci sit near genes with a plausible role in fungal defense rather than being statistical passengers of population structure.
02 Molecular Biology and Evolution · 18 September 2026_
Uncovering structural variation in conifer gigagenomes: evolutionary insights and technical challenges
- Structural variation
- Methods
- Population genetics
Long-read sequencing of three hybridizing pine species finds 5.7 million structural variants, then shows that graph-pangenome genotyping from short reads recovers only 44% of them in these repeat-rich genomes.
Summary
Conifers have large, repeat-rich genomes that have made structural variant (SV) discovery difficult. The authors combined long- and short-read sequencing across Pinus densata, P. tabuliformis and P. yunnanensis - three closely related pines with a hybridization history. From 21 long-read genomes they identified 5.7 million SVs, mostly in intergenic and intronic regions and heavily associated with transposable elements, with loss-of-function mutations hundreds-fold more common among SVs than SNPs. They then tested whether a graph pangenome built from the long-read SVs could genotype those same variants from 29 short-read genomes: recall was 44% at 70% precision. Selection scans run separately on SNPs and on genotyped SVs agreed on only 19% of candidate loci.
What to take from it
- 21 long-read genomes across three pine species: 5.7M SVs (52% insertions, 43% deletions, remainder inversions/duplications/translocations)
- 97% of SVs sit in intergenic or intronic regions; 60% overlap transposable elements
- Loss-of-function mutations are hundreds-fold more common among SVs than among SNPs
- Graph-pangenome SV genotyping from short reads in 29 individuals: 44% recall at 70% precision
- SNP-based and SV-based selection scans agreed on only 19% of candidate gene loci
Review
The useful result here is the cautionary one: graph-pangenome genotyping recall of 44% in conifers is well below what is typically reported for human or crop pangenomes, a direct consequence of gigagenome repeat content, and it is a concrete number anyone applying standard graph-genotyping pipelines to large, repetitive genomes should benchmark against before trusting their own recall. The low overlap between SNP-based and SV-based selection scans (19%) is a second caution: which marker type is used changes which genes look like they are under selection, so single-marker-type adaptation stories in repeat-heavy genomes should be treated skeptically until corroborated by the other marker type. The population diversity concordance between SVs and SNPs is reassuring for broad-strokes demography, but the paper is explicit that fine-scale SV genotyping accuracy is the current bottleneck, not the biology.
03 Genome Biology · 16 September 2026_
dicast: a machine learning method for accurate structural variant detection from short-read sequencing data
- Methods
- Structural variation
- Human genetics
A machine-learning scorer trained on a manually curated, multi-technology ground truth from nine samples outperforms existing short-read structural variant callers and finds 20% more candidate pathogenic deletions in diagnostic cohorts.
Summary
Structural variants cause human disease but remain hard to detect from short-read sequencing, the technology behind most clinical workflows. The authors built a new ground-truth SV set from nine samples across multiple sequencing technologies, with extensive manual curation, and trained dicast, a machine-learning method that scores SV calls using alignment and genomic-context features. dicast outperforms existing short-read callers and consensus approaches, recovers more true positives at matched precision, identifies all pathogenic variants in tested disease cohorts, and calls 20% more candidate pathogenic deletions than consensus approaches.
What to take from it
- New multi-technology ground-truth SV set built from nine samples, manually curated
- dicast: a trained ML classifier that scores SV calls from short-read alignment and genomic-context features
- Outperforms existing short-read SV callers and consensus approaches at matched precision
- Recovers all pathogenic variants in tested disease cohorts, plus 20% more candidate pathogenic deletions than consensus methods
Review
Because short reads remain the clinical default, a scoring method that closes part of the SV-detection gap without requiring long-read sequencing has direct diagnostic relevance. The ground truth behind it is built from only nine samples, so how the learned model generalizes to SV classes, genomic contexts or sequencing chemistries not represented in that training set is the open question - the usual failure mode for an ML-based caller is distribution shift away from its training data. The 20% increase in candidate pathogenic deletions is a call-count claim, not a confirmed-diagnosis claim: those additional candidates need orthogonal validation, ideally long-read or functional, before they change patient management.
04 Journal of Dairy Science · 17 September 2026_
Genomic background of gestation length and calving-related traits in Holstein cattle
- Animal science
- Population genetics
A single-step GWAS across 226,717 Holstein cows with a 461,819-animal pedigree finds calving-related traits are mostly polygenic, with one genome-wide significant hit for calving ease near FARS2 on BTA23.
Summary
Calving-related traits in dairy cattle - calf size, calving ease, gestation length and stillbirth - generally have low heritability but enough additive genetic variance for genomic selection to work. The authors estimated genetic parameters and ran single-step GWAS using 226,717 phenotyped animals, 15,761 genotyped with 45,101 SNPs, and a 461,819-animal pedigree. Heritability ranged from 0.001 (gestation length in heifers) to 0.16 (gestation length in cows), with genetic correlations up to 0.74 between gestation length measured in heifers versus cows. Most GWAS signals fell short of genome-wide significance; the one that cleared it was calving ease in cows on BTA23, flagging FARS2 as a positional candidate gene, alongside several other suggestive candidates (NPPA, NPPB, BCHE, EPHA4, DLD, GTF2I) at other loci.
What to take from it
- 226,717 phenotyped animals, 15,761 genotyped with 45,101 SNPs, 461,819 animals in the pedigree
- Heritability from 0.001 (gestation length, heifers) to 0.16 (gestation length, cows)
- Genetic correlation up to 0.74 between gestation length in heifers and cows; -0.57 between calving ease and stillbirth in heifers
- One genome-wide significant hit: calving ease in cows, BTA23, candidate gene FARS2
- Architecture described as predominantly polygenic, with most signals below genome-wide significance
Review
The scale (226,717 phenotyped animals) is what makes the heritability and genetic-correlation estimates trustworthy, but the GWAS itself is thin: only one signal clears genome-wide significance across four traits, and the paper explicitly frames the architecture as predominantly polygenic. That means a polygenic score built from many small-effect loci, not single-gene diagnostics, is the realistic path to genetic gain here. FARS2 as a candidate for calving ease is positional, not functionally confirmed - it sits near the significant SNP. Given heritability as low as 0.001 for some traits, genetic progress on those specific measures will be slow regardless of how well the underlying loci are eventually mapped.
05 Frontiers in Genetics · 3 September 2026_
Influence of homozygosity on genomic structural variation analyses for predicting ACL rupture risk in the Labrador Retriever and Rottweiler
- Structural variation
- Animal science
Breed-specific structural-variant maps and inbreeding measures across 1,166 Labradors and Rottweilers find that inbreeding coefficient predicts ACL rupture risk in Rottweilers but not in Labradors.
Summary
Anterior cruciate ligament (ACL) rupture is a common orthopaedic disease in dogs with breed-specific prevalence. The authors used breed-specific de novo genome assemblies to characterize structural variation between Labrador Retrievers and Rottweilers, then quantified runs of homozygosity and tested their association with ACL rupture risk in 1,058 Labradors (464 cases, 594 controls) and 108 Rottweilers (83 cases, 25 controls). Homozygosity measures were not significantly associated with rupture risk in Labradors, but inbreeding coefficient significantly raised risk in Rottweilers (odds ratio 2.01, 95% CI 1.12-3.98). A combined-breed model found multicollinearity between breed identity and homozygosity content, meaning the two factors cannot be cleanly separated statistically.
What to take from it
- 1,058 Labrador Retrievers (464 cases/594 controls) and 108 Rottweilers (83 cases/25 controls)
- Breed-specific de novo genome assemblies used to characterize structural variation between the two breeds
- Inbreeding coefficient significantly raised ACL rupture risk in Rottweilers (OR 2.01, 95% CI 1.12-3.98) but not in Labradors
- Combined-breed analysis found multicollinearity between breed and homozygosity content, so the two cannot be statistically disentangled
Review
The Rottweiler sample is small and imbalanced (83 cases versus 25 controls), so the reported odds ratio carries wide uncertainty, and the authors themselves call for independent validation in a larger cohort before treating this as a usable risk marker. The breed-specific result - inbreeding predicts risk in one breed but not the other - is a genuine negative finding against assuming risk factors generalize across breeds, but it also means the underlying causal structural variants, rather than genome-wide homozygosity as a proxy, still have not been identified. This reads as evidence that within-breed inbreeding avoidance is worth watching in Rottweilers specifically, not as a validated genetic test ready for breeder use.