01 Frontiers in Plant Science · 17 September 2026_
Pangenomic characterization and grain-development expression analysis of the OVATE gene family in sorghum
- Structural variation
Across 17 sorghum genomes, 498 OVATE family genes fall into 35 orthogroups, of which only 5 are core; two genes dominate expression in the developing grain.
Summary
OVATE family proteins regulate organ shape and grain development. The authors annotated them in 17 published sorghum assemblies and found 498 genes in 35 orthogroups: 5 core, 9 softcore and 21 shell by presence/absence. Core genes are present in every accession and more conserved; most of the family is under purifying selection, with elevated Ka/Ks in some pairwise comparisons for OFP7, OFP13, OFP25, OFP26 and OFP30. Expression is concentrated in seed and inflorescence. Hormone treatments (ABA, BR, MeJA, IAA) change expression roughly as the promoter cis-elements predict. SbOFP14 and SbOFP15 dominate the grain transcriptome, and co-expression networks link hub genes SbOFP32 and SbOFP13 with NAC, MYB, MADS and bHLH factors.
What to take from it
- Most OVATE orthogroups are shell, so a single sorghum reference misses much of the family
- SbOFP14 and SbOFP15 are the grain-expressed candidates
- Hormone responses match promoter elements, but this is correlation only
Review
This is a gene-family pangenome survey: useful as an inventory, light on causation. The core/shell split depends on annotation, and 17 assemblies from different groups and pipelines will differ in gene calling; a gene missing from one annotation is not necessarily absent from the genome. The paper should show PAV calls confirmed by read mapping or synteny-aware search against each assembly. Elevated Ka/Ks 'in subsets of comparisons' is weak evidence of selection without a codon model and multiple-testing control. No variant is tied to a grain trait, and there is no knockout or association test. Before using these genes as breeding targets, ask for a PAV-trait association in a diversity panel and functional evidence for SbOFP14 or SbOFP15.
02 Genome · 3 October 2026_
Integration of cytogenetics and genomics reveals translocation and recombination between nonhomologous chromosomes in a novel wheat line with dual introgression
- Structural variation
A stable wheat line from a Trititrigia x triticale cross carries Thinopyrum and rye segments on three wheat chromosomes, mapped by FISH, GISH and short-read coverage.
Summary
Line 13-26 (2n=42) was derived from octoploid Trititrigia crossed with hexaploid triticale and selfed for 17 generations. It has heavier kernels than both parents and more protein than the male parent. Sequential GISH, multicolour FISH and short-read sequencing show about 220 Mb missing from wheat 6AS and 89 Mb from 2DL, replaced by about 221 Mb of Thinopyrum intermedium 6JvsS and 42 Mb of rye 5RL. The karyotype is T6JvsS.6AL plus terminal rye 5RL segments on 2DL and 4BL, so the line carries translocations between non-homologous chromosomes from two donor genera.
What to take from it
- Read coverage against the wheat reference sizes alien segments to tens of Mb
- Rye 5RL is joined to two non-homologous wheat arms
- Cytogenetics and sequencing together detect segments either method alone would miss
Review
The karyotype is well supported: two cytogenetic methods and sequencing agree, and segment sizes come from data rather than banding. The limits are resolution and agronomy. Short-read coverage against Chinese Spring locates deletions but cannot resolve breakpoints at the junctions, which are the interesting part for non-homologous recombination; long reads or an assembly of the line would. Agronomic superiority is stated for kernel weight and protein but the abstract gives no replicated field trial or environments. Rye 5RL on two wheat arms raises the question of stability and transmission, which 17 selfing generations suggest but do not measure. Before using 13-26 in breeding, ask for breakpoint sequences, multi-site trials and markers for each alien segment.
03 Current Research in Microbial Sciences · 29 August 2026_
Within-patient carbapenem heteroresistance in klebsiella pneumoniae associated with a repeat-associated blaNDM-1 resistance-module deletion
- Structural variation
- Microbial genomics
Resistant and susceptible K. pneumoniae from one wound differ only by a 10.5 kb plasmid deletion between two identical sul1-qacEdelta1 repeats carrying blaNDM-1.
Summary
Three ST37/KL104 isolates came from one hospitalised patient: KP02 and KP03 carbapenem-resistant, KP04 susceptible, with KP03 and KP04 from the same wound culture. Three colonies per isolate were sequenced; there were no differing SNPs in the shared core. The difference is a 10,483 bp segment on an IncHI1B/IncFIB(K) plasmid, bounded by two identical direct repeats containing sul1 and qacEdelta1. Read-level checks, an independent assembly and structural PCR support a repeat-mediated deletion. An outward-facing PCR product suggests a circular intermediate, not fully resolved. Without antibiotics the isolate lacking the module grew faster and to higher biomass.
What to take from it
- Identical flanking repeats let a resistance module be lost in one step
- Resistant and susceptible states can coexist in one specimen with no SNP difference
- Short-read SNP typing would call these isolates identical
Review
The structural evidence is good for a single case: colony replicates, independent assembly and PCR all point the same way. It is one patient and one plasmid, so the frequency of this event is unknown. The growth advantage is shown without antibiotics but, as the authors say, not tied to the deleted module itself; a complementation or reinsertion experiment would settle it. The circular intermediate is only inferred. For clinical labs the practical point is that heteroresistance from plasmid rearrangement needs long reads to see; before generalising, ask for a cohort screen of NDM plasmids for the same flanking repeats and a measure of how often susceptible subpopulations appear.
04 Nature Communications · 2 September 2026_
Common alleles associated with connectivity within and across the brain's main functional networks
- Human genetics
- Methods
A multivariate GWAS of resting-state connectivity finds 114 independent SNPs and 315 candidate genes, with shared and network-specific signals.
Summary
Brain functional connectivity is a high-dimensional phenotype, and univariate GWAS of single network edges lose power. The authors run a multivariate GWAS across connectivity measures within and between the main functional networks. This yields 114 independent significant SNPs pointing to 315 candidate genes. Much of the genetic signal is shared across networks, with some network-specific loci separating cognitive from sensory-motor networks. Connectivity loci overlap substantially with neuropsychiatric risk loci.
What to take from it
- Multivariate testing raises discovery for correlated imaging traits
- Cognitive and sensory-motor networks show partly separate genetic signals
- Connectivity loci are enriched for neuropsychiatric risk SNPs
Review
The method choice is sound for a correlated multi-trait phenotype, and the transfer to livestock or plant multi-trait GWAS is direct. The weaknesses are interpretive. Multivariate hits say a SNP affects some combination of traits, not which; network-specific claims need conditional or colocalisation tests per trait. Resting-state connectivity is sensitive to head motion and scanner, and these covariates can carry genetic signal of their own. 'Large genetic overlap' with psychiatric disorders depends on the method (genetic correlation, conjFDR, enrichment), and overlap does not show shared causal variants. Before using these loci, ask for replication in an independent imaging cohort, the per-trait effect pattern for each lead SNP, and fine-mapping that resolves the 315 genes to fewer.
05 Nature Communications · 2 September 2026_
End-to-end deep learning methods for genetic risk prediction of schizophrenia
- Human genetics
- Methods
Neural networks on exome data from 6,135 cases and 6,245 controls predict schizophrenia better than additive models once the sample is large enough.
Summary
The authors train genome interpretation neural networks end to end on whole-exome sequencing of 6,135 schizophrenia cases and 6,245 controls and compare them with additive models. Nonlinear networks outperform additive ones when sample size is sufficient. Explainable-AI methods and biological priors are used to identify predictive genes and pathways; known risk genes are recovered and BASP1, involved in neuronal development, is proposed as an understudied candidate.
What to take from it
- Nonlinear models beat additive ones only above a sample-size threshold
- Exome-only input leaves out most common-variant signal
- BASP1 is a candidate from model attribution, not from association
Review
The comparison is useful because it asks when nonlinear models pay off rather than assuming they do. The weak point is the leap to epistasis: better prediction from a network can come from better handling of rare-variant burden, population structure or batch effects between case and control sequencing, not only from interactions. The paper needs to show the gain survives ancestry and batch correction and that a model with explicit pairwise terms explains it. Twelve thousand samples is small for deep learning, and there is no external validation cohort in the abstract. A fair baseline is a modern PRS plus rare-variant burden, not an additive model on the same exome. For livestock genomic prediction, where nonlinear models are also tested, this is a reminder to demand out-of-population validation before trusting gains.