01 Research Squarepreprint · 8 September 2026_
Z-DNA-associated genomic instability in the human pangenome
- Structural variation
- Human genetics
Predicted Z-DNA-forming sequences across T2T-CHM13 and 464 HPRC haplotypes carry more indels and complex variants than matched controls, most near B-Z junctions.
Summary
Predicted Z-DNA-forming sequences were mapped in T2T-CHM13 and 464 haplotype-resolved HPRC assemblies. Density was similar across haplotypes and superpopulations but differed between CHM13 and GRCh38 in repetitive compartments, including acrocentric regions. Across 52 million HPRC variants, these loci had higher mutation density than controls matched for length, GC and CpG/GpC content, most strongly for small and intermediate indels and complex insertion- or deletion-like events. Enrichment was highest near predicted B-Z junctions and in genic and euchromatic regions. The same mutation classes were enriched among more than 250,000 de novo mutations.
What to take from it
- Z-DNA motif density is stable across human haplotypes; the CHM13 versus GRCh38 difference is a reference-completeness effect
- Indels and complex variants, not SNVs, show the strongest enrichment
- De novo mutations show the same pattern, which argues against a pure calling artefact
- The motif catalogue is a reusable annotation for SV and indel calls on HPRC graphs
Review
Using the HPRC assemblies, not one reference, is what makes the per-haplotype density comparison possible, and the matched controls are the right design. The weak point is that every Z-DNA locus is predicted from sequence; none is shown to adopt Z conformation in cells, so the result is about a sequence class, which is also a class of alternating purine-pyrimidine repeats. Repeats are where assembly, alignment and variant calling all fail most, and higher indel density there is expected from error alone. Before using it, ask for the enrichment stratified by tandem-repeat annotation and by caller or graph decomposition method, and for confirmation in de novo calls made from long-read trios. This is a preprint and not peer reviewed.
02 bioRxivpreprint · 24 September 2026_
Inversion characterization in Timema stick insects: from local PCA to pangenome approaches
- Structural variation
- Population genetics
- Methods
Local PCA on GBS data, pairwise SyRI alignments of eight assemblies and an eight-genome pangenome graph find the same inversion regions but disagree on small and complex ones.
Summary
Inversions in Timema cristinae were characterised three ways: local population structure from genotyping-by-sequencing data, SyRI calls from pairwise alignments of eight de novo assemblies, and a pangenome built from the same eight assemblies. All three find the broad regions carrying SVs. Local PCA detects only the largest inversions. Pairwise alignment finds the most inversions, but results depend on how shared inversions are matched across genome pairs. The pangenome scales and represents complex SVs as graph structures, but needs different visualisation and interpretation.
What to take from it
- Local PCA on reduced-representation data finds only large inversions
- Pairwise SyRI calls give the highest count but need a rule for merging across pairs
- A pangenome graph avoids the pairwise merge problem and represents nested or complex SVs
- Data type and method should be chosen by the size of inversion that matters
Review
A useful side-by-side on one system, and the trade-offs it reports match what livestock and plant groups see when moving from SNP-array scans to multi-assembly graphs. It is not a benchmark: there is no truth set, so "detects more" cannot be separated from "calls more false positives". Eight assemblies is small for a pangenome, and the abstract names neither the graph builder nor how inversions were extracted from bubbles, both of which change the result. Before generalising, ask for validation of a subset by long reads or Hi-C, per-size-class concordance between methods, and the graph pipeline parameters. Preprint, not peer reviewed.
03 Mammalian Genome · 29 September 2026_
Development and validation of a high density SNP array for indigenous Indian pigs
- Animal science
- Population genetics
Resequencing of eight Indian pig populations gave 61 million variants, filtered to a 635,974-SNP array with 99.6% call rate on 96 test samples.
Summary
Whole-genome resequencing of eight indigenous Indian pig populations found more than 61 million variants. These were filtered to 635,974 bi-allelic SNPs with genome-wide coverage for a custom array. Genotyping 96 pigs from several breeds gave a 99.6% average call rate; 553,289 markers were polymorphic and 0.8% monomorphic, and 579,504 were classed as best and recommended. The array resolved structure among Indian pigs, and in silico extraction of its SNPs from whole genomes separated indigenous and global breeds.
What to take from it
- First high-density array designed from Indian pig diversity rather than commercial breeds
- Technical performance is high: 99.6% call rate, under 1% monomorphic markers
- Separates indigenous from global breeds, so suited to admixture and conservation work
- Commercial arrays ascertained in European breeds under-represent this variation
Review
The design solves a real ascertainment problem: arrays built from European commercial breeds miss variation in local populations. Validation is technical only. Ninety-six samples show call rate and polymorphism, not imputation accuracy to sequence, genomic prediction accuracy or GWAS power, which are what a breeding programme needs. The discovery panel of eight populations, with sample sizes not given in the abstract, sets which SNPs pass the frequency filter, so populations outside it may still be under-represented. Ask for imputation accuracy against held-out sequence, per-population minor allele frequency spectra, and a comparison with the existing commercial porcine arrays on the same animals.
04 medRxivpreprint · 10 September 2026_
Yield of long-read genome sequencing for rare disease diagnosis in short-read genome negative cases
- Structural variation
- Human genetics
Long reads on 107 short-read-negative rare disease cases gave 13 new diagnoses, five undetectable by short reads: a 4.7% incremental yield.
Summary
144 cases from a GREGoR centre that already had short-read genomes were resequenced with long reads; 107 were undiagnosed. Thirteen received new diagnoses, five of which short reads could not have found, an incremental yield of 4.7%. Those five came from variants in short-read dark regions, structural variants, tandem repeat expansions, and de novo calls made with only one parent available. Two needed RNA-seq to confirm aberrant transcripts, and one was supported by methylation episignature from the long-read data.
What to take from it
- Strict incremental yield over short-read genomes is 4.7%, not the 13 total new diagnoses
- Gains come from dark regions, SVs, repeat expansions and phasing
- One assay gives sequence and methylation, which supported one diagnosis
- RNA-seq was still needed to confirm two of five
Review
The paired design is the strength: same patients, short-read negative first, so the 4.7% is a true increment and not mixed with reanalysis gains (the other eight diagnoses). The figure is from 107 cases at one centre, so its confidence interval is wide, roughly 1.5 to 11%, and depends on how much the short-read pipeline already did for SVs and repeats. Cost per added diagnosis is not given. Before using 4.7% for planning, ask for the short-read pipeline details, the interval on the yield, and replication in other GREGoR centres. Preprint, not peer reviewed.
05 Nature Communications · 29 August 2026_
A pre-train and fine-tune framework for adaptive boosting of pre-trained polygenic risk scores
- Methods
- Human genetics
Fine-tunes an existing polygenic score by adaptive variable selection and boosting on new data, with FDR control on the added signal; validated in All of Us, eMERGE and Penn Medicine Biobank.
Summary
The method takes a pre-trained polygenic score and refines it with adaptive variable selection and boosting, looking for predictive signal the original model missed. Simulations show it finds signal orthogonal to the pre-trained score while controlling the false discovery rate. Scores for binary diseases and continuous traits were fine-tuned in UK Biobank and validated in All of Us, eMERGE and Penn Medicine Biobank, with significant gains in some settings and similar performance in others.
What to take from it
- Treats a published PRS as a starting point and adds signal, instead of retraining from scratch
- FDR control limits added variants to signal orthogonal to the base score
- Validated in three external biobanks
- Gains are significant in some traits and absent in others
Review
The idea transfers directly to livestock, where a published genomic prediction can be adapted to a new population without the full training data. The abstract reports "significant improvements in several scenarios" without effect sizes, so the gain may be small in absolute R2 or AUC. Fine-tuning in UK Biobank risks reusing samples that trained the base scores, which would inflate gains; external validation addresses this only if those cohorts were not in the base GWAS. Before adopting it, check the absolute improvement per trait, the overlap between fine-tuning data and base-score training data, and performance in non-European ancestries.