CIRI3 detects and quantifies circular RNAs in terabyte-scale RNA-seq datasets, running an order of magnitude faster than existing tools with increased accuracy through dynamic multithreaded task partitioning and a blocking search strategy for junction reads. Applied to 2,535 cancer-related samples, it identified differentially spliced circRNAs and yielded the CIRIonco database with a pretraining model and biomarker network.
A tractography-derived measure of brain-wide tumor involvement, the lesion–tract density index (L-TDI), outperformed conventional tumor volume for predicting overall survival in 999 patients with newly diagnosed glioblastoma across multiple institutions. Tumor volume's prognostic value depended strongly on analytical design and was largely confined to the contrast-enhancing compartment, whereas L-TDI showed robust associations with overall survival across thresholds and tumor compartments and retained its value after adjustment for established clinical variables. The findings support assessing glioblastoma prognosis as a brain-wide disease rather than by local tumor extent.
Cross-source progression assessment in advanced pancreatic cancer frequently failed to yield immediate confirmation of progression. Among 645 patients in the primary estimand, only 25 (3.88%) showed same-day concordance, 219 (33.95%) confirmation, 108 (16.74%) explicit nonprogression, and 166 (25.74%) were indeterminate. Among classifiable first opportunities, 33.03% documented explicit nonprogression, supporting an opportunity-adjusted six-state framework over binary discordance estimates.
AdaptiveFlow, an open-source platform for ultralarge virtual screening, offers a ready-to-dock version of Enamine REAL Space comprising 69 billion compounds and integrates more than 1,500 docking protocols, scaling near-linearly on up to 5.6 million CPUs. Screening with an 18-dimensional property grid and optional active learning cut computational costs by orders of magnitude and yielded nanomolar inhibitors of FSP1 and PARP1, with co-crystal structures explaining FSP1 inhibition. It lowers the cost and scalability barriers that have limited billion-scale screening.
Not routinely assessing gastric residual volume in ventilated critically ill children was noninferior to 6-hourly assessments for survival and ventilator-free days at 30 days (median 25 days in both groups; adjusted OR 0.95 [95% CI 0.86-1.05]). In the trial of 4700 children in pediatric intensive care units in the UK and Switzerland, dropping routine GRV checks also improved feeding: mean energy requirements achieved by 72 hours were 80.3% versus 76.8% (adjusted mean difference 3.2 percentage points, P < .001).
GPN-Star, a phylogeny-aware genomic language model that uses whole-genome alignments and species trees, outperforms previous methods across variant effect prediction tasks in coding and non-coding human genome regions. Trained on vertebrate, mammal and primate alignments, it improves prioritization of pathogenic and fine-mapped GWAS variants, yields strong complex trait heritability enrichments, and improves rare variant association testing power. Versions trained for five model organisms, from mouse to Arabidopsis, show the framework generalizes.
A blood drying protocol with thermal treatments and primer-limited reverse transcription stabilizes RNA in whole blood, allowing direct amplification without extraction, purification, or cold-chain logistics. Detection reached 10 copies per microliter for Zika virus and 1 international unit per microliter for HCV, and larger blood volumes achieved single-copy sensitivity without sample loss. Zika and HCV were detected with a portable fluorometer, supporting field-deployable point-of-care diagnostics.
The paper describes an engineered nanopore that identifies saccharides, amino acids, peptides and ribonucleotides. Machine learning classifiers, including a general classifier and few-shot learning approaches, were trained to identify these molecule classes from nanopore signals. The data and custom code are provided with the paper.
Peripheral-blood cfDNA matched or exceeded marrow-plasma cfDNA as a liquid biopsy for multiple myeloma, despite the tumor's marrow confinement. In 74 patients profiled across 372 cfDNA time points, blood cfDNA recovered marrow-defined CNAs and mutations at 77.4% vs 72.0% and 69.2% vs 65.4%, and at high tumor fraction adverse-lesion sensitivity reached 100% in both compartments. Marrow cfDNA did show distinct, handling-sensitive fragmentomic architecture (LOOCV AUC=0.910), so compartment, biomarker class, tumor fraction and processing all shape liquid-biopsy performance.
Why it matters: It shows a simpler blood draw is not inferior to marrow-derived cfDNA for myeloma genomic profiling, with practical processing caveats.
Residual intrafraction motion during gated MR-guided prostate SBRT with simultaneous integrated boost caused directionally asymmetric coverage loss, hitting the boosted GTV hardest: per-fraction median GTV deviations were −1.2% for D95 and −5.6% for V45Gy across 150 fractions in 30 patients. Motion-inclusive dose reconstruction yielded asymmetric internal target volume margins (2/2/1/2/1/2 mm for GTV) that achieved sufficient coverage in about 94% of fractions per direction, at the cost of an average 6.9 ± 4.7% reduction in gating duty cycle.
Why it matters: Provides concrete population-based margins that trade a small loss of delivery efficiency for robust GTV coverage in gated MR-guided prostate SBRT.
DeepSomatic, a deep-learning method, detects somatic small nucleotide variations and insertions and deletions from both short-read and long-read data, with modes for whole-genome and whole-exome sequencing and support for tumor-normal, tumor-only and FFPE samples. To enable training and benchmarking, the authors generated the openly available CASTLE dataset of six matched tumor-normal cell line pairs sequenced with Illumina, PacBio HiFi and Oxford Nanopore Technologies. Across cell line and patient-derived samples and across sequencing technologies, DeepSomatic consistently outperforms existing callers.
GUTchetp predicts gut microbial biotransformation products using an ensemble of monolingual and multilingual neural machine translation models, reaching a Top-20 BLEU score of 66.59% and MaxFrag accuracy of 67.19% on a benchmark of six compound categories. A second component identifies associated microbial enzymes and species, tested on 95 metabolism events, and raises accuracy by 34.38 percentage points over existing tools via a re-ranking rule combining EC-classifier consistency with reaction similarity. Together the components capture 68.42% of known gut microbial biotransformation profiles, outperforming prior computational tools.
Patient-specific proliferation and diffusion rates of a reaction-diffusion model can be inferred from a single routine H&E biopsy by fitting cell-nucleus point patterns, validated across eleven multicenter cohorts. The derived dispersion length and front velocity stratified progression-free and overall survival independent of routine covariates. A virtual glioblastoma trial allocating hypofractionation to faster-growing tumors and dose intensification to slower-growing ones added 96.7 days of restricted mean progression-free survival over standard of care (permutation p < 0.001).
Why it matters: It turns a single routine H&E biopsy into patient-specific tumor growth parameters usable for survival prediction and virtual treatment trials.
DIAL is a modular framework for building editable promoters that tune transgene expression in fine, heritable steps. Recombinase-mediated excision of nested spacers between zinc finger transcription factor binding sites and the core promoter generates a tunable range of unimodal expression setpoints from a single promoter, with small-molecule temporal control. Lentiviral delivery produced multiple setpoints in primary cells and induced pluripotent stem cells, and enabled mapping of transgene levels to phenotype in direct conversion to induced motor neurons.
Rationally engineered chimeric attachment sites boost Bxb1-mediated integration efficiency in human cells and plants. The attB(V111) variant achieved 51.9% integration in HEK293T cells (1.7-fold over wild-type attB) and 35.6% in rice protoplasts (4.4-fold), and paired with a Bxb1 catalytic-domain mutant delivered 31% integration of a CD19 CAR cassette and 25% of an ornithine transcarbamylase cassette in human cells. In rice, a 5.8 kb herbicide-resistance cassette integrated in 24% of regenerated plants, with heritable transmission to T1 plants.
Sea nettle jellyfish venom kills human cells through apoptosis and necroptosis, and whole-genome CRISPR screening identifies proteoglycan biosynthesis as the key host pathway governing venom action. Heparin blocks venom cytotoxicity at a clinically relevant dose, works therapeutically even 1 hour after exposure, and in vivo reduces venom-induced acute spontaneous pain, thermal hyperalgesia, and mechanical allodynia. The findings point to repurposing heparin as a prophylactic or therapeutic against sea nettle stings, for which no rational treatments exist.
An industrialized surface protein proximity-mapping workflow produced 248 proximity maps across 12 receptor tyrosine kinases and 28 tumour cell systems, yielding MetaMap, a framework that defines spatial protein communities and the concept of tumour-associated proximity antigens (TAPAs). Applied within a multimodal prioritization framework, it identified EGFR-CDCP1 as a TAA-TAPA pair that enhances tumour cell killing across therapeutic modalities, making disease-specific membrane proximity a design principle for precision multispecific therapeutics.
Pyrroles can now be made directly from isoxazoles by an O-to-C skeletal edit that swaps the single differing ring atom in a one-pot sequence. The key intermediate is an N-propargylic enaminone, giving a retrosynthetic disconnection orthogonal to traditional pyrrole syntheses. A predictive computational model captures the conformational features controlling the reaction outcome, enabling regioselective access to otherwise elusive pyrrole scaffolds.
CRISP, a vision-based pathology foundation model trained exclusively on frozen section slides, outperformed current foundation models and showed extensive validation, including a prospective cohort, for supporting treatment decisions across the surgical workflow. It brings intraoperative pathology decision support closer to clinical use.
A suite of large pre-trained language models extracts computable clinical data from unstructured electronic health records and integrates the findings into knowledge graphs, alongside structured EHR data, to build longitudinal patient journeys. The output is intended to help understand patient trajectories and treatment responses in the real world. The work was published in Nature Medicine on 10 September 2026.
