Articles

The study generated high-quality phased diploid genome assemblies for 10 Japanese and 9 Saudi individuals using PacBio HiFi, ultra-long Oxford Nanopore, and Hi-C sequencing, and constructed population-specific pangenome graphs as well as a population merged graph with standard reference backbones, using Minigraph-Cactus pipeline. Analysis showed that aligning short reads to these population-tailored graphs improved variant detection, particularly for structural variants, compared to conventional references or mismatched pangenomes. The graph-based approach also captured complex copy number variation at CYP2D6 locus.

There are four main human reference genomes: GRCh37, GRCh38, CHM13, and the pangenome. GRCh37 was the first developed, and many clinical labs still use it due to its familiarity and the cost of change. GRCh38 is the most widely used, due to its rich annotation. CHM13 is the newest and most complete linear human reference. The pangenome accounts for population diversity and is already integrated into DRAGEN. The current VCF format may not fully capture pangenome reference information. Liftover methods can reannotate variants on updated references, but they may lead to variant loss. In the future, new AI-based approaches may enable annotation to follow de novo assembly directly.

Snow leopards have the lowest genetic diversity of any big cat species, likely due to persistently small population size thrughout their evlutionary history rather than recent inbreeding. Snow leopards also have significantly less highly deleterious homozygous load compared to other Panthera species, suggesting effective purging during their evolutionary history at small population sizes.

A Swiss Tumor Profiler study evaluated the feasibility of using whole-genome sequencing (WGS) on FFPE metastatic melanoma samples for routine molecular pathology. DNA extraction: RecoverAll Total Nucleic Acid Isolation Kit. Library Prep: TruSeq DNA Nano. 150 bp paired-end reads NovaSeq 6000. Average coverage 73× and 38× for tumor and normal. Secondary analysis with DRAGEN v3.9.5. Result compared with commercial targeted panels.

Using F1CDx assay as truth set, WGS identified 38% more clinically actionable markers than FoundationOne CDx and 71% more than MelArray. It showed high sensitivity for short variants (SNVs ~95%, MNVs ~98%, indels ~90%) and homozygous deletions (~96%), though lower for amplifications (~76%). TMB measurements strongly correlated with F1CDx (R = 0.98), particularly at the ≥10 mutations/Mb clinical cutoff. Sensitivity drops for very low–allele frequency variants (<5%) and in highly degraded or low-purity samples.

A UK Biobank WGS study on 490K participants, of which 450K are European ancestry. The sequencing was 32.5x on average, using NovaSeq 6000. Variant calling was performed with DRAGEN v3.7.8, and joint calling with GraphTyper. The study found 1.5 billion variants, which is an 18.8-fold increase over genotyping with imputation, and over a 40-fold increase compared to WES. It also improved the detection of SVs, rare UTR, and non-coding variants. The study identified 28K genome-wide significant associations, of which 26K were observed in individuals of European ancestry, and 82 were observed only in one of the non-European analyses.

81% of tropical tree species depend on animals for seed dispersal. Declines in animal diversity and movement severely disrupt seed dispersal. Areas with the least seed dispersal disruption had four times the aboveground carbon accumulation compared to those with the most disruption. Animal conservation need to be integrated with climate mitigation strategies.

A metabarcoding study of four decades of samples from Germany’s Environmental Specimen Bank. Although no consistent loss in species richness was observed, a gradual and widespread shift in species composition across both terrestrial and aquatic ecosystems was detected. DNA was extracted from mixed-organism samples and analyzed using metabarcoding. Sequencing was performed with MiSeq paired-end 500- and 600-cycle kits, and the reads were clustered into taxonomic units. These were then matched against reference databases for species identification. The study focused on presence–absence data and changes in community composition.

The study conducted eight genome-wide association analyses of self-reported stuttering, stratified by sex and ancestry, using a sample of 0.1 million cases and 1.0 million controls, derived from 23andMe genotype data. It identified 57 unique loci associated with stuttering. The heritability of stuttering was found to be in line with SNP-based estimates for other common, complex traits. Genetic correlation analysis was performed to identify shared genetic architecture between stuttering and related traits. Mendelian randomization was used to test whether stuttering and comorbid traits have a causal relationship. These analyses revealed genetic similarities between stuttering and traits such as autism, depression, and impaired musical rhythm across sexes. The PRS model was tested on independent datasets, yielding an AUC of less than 0.6.

Freshwater fish show a preference for textured concrete panels that mimic natural roots, compared to flat control panels, under test tank simulations. Similar patterns have been observed previously in saltwater fish. Integrating engineered roughened structures into waterways may enhance habitat complexity and boost biodiversity.

A deep learning model that predicts the regulatory effects of genetic variants by analyzing 1 megabase of DNA sequence at single-base resolution. It is trained on large-scale datasets from both human and mouse genomes across multiple cell types. The model simultaneously captures diverse genomic signals such as gene expression, RNA splicing, chromatin accessibility, and 3D genome contacts. It is particularly effective for multimodal variant effect prediction in a single inference. A major limitation is its restricted input length of 1 megabase.

A hybrid variant calling approach using approximately 15x short-read (Illumina) and 15x long-read (Nanopore) data, processed with a retrained DeepVariant model, improves variant calling accuracy compared to using either technology alone (30x for both short and long read). This method is particularly effective in complex genomic regions such as tandem repeats and low-mappability areas. The SNV F1 score reached 0.9980, and the indel F1 score was 0.9454.