Search

A technique called condense-seq has been developed to measure nucleosome condensability and used to show that mononucleosomes contain sufficient information to condense into large-scale compartments without requiring any external factors.

A deep learning model, WaveCastNet, predicts seismic ground motion in real time. It outperforms traditional methods by accurately forecasting wave behavior, even in rare and large earthquakes, without relying on magnitude or epicenter location data

Human decision confidence displays a number of biases and has been shown to dissociate from decision accuracy. Here, by using neural network and Bayesian models, the authors show that these effects can be explained by the statistics of sensory inputs.

Identifying jets originating from heavy quarks plays a fundamental role in hadronic collider experiments. In this work, the ATLAS Collaboration describes and tests a transformer-based neural network architecture for jet flavour tagging based on low-level input and physics-inspired constraints.

Vinyard et al. present a generative method to model cell dynamics using neural stochastic differential equations that learn state-dependent drift and diffusion, outperforming existing approaches and enabling perturbation studies of development and disease.

Non-equilibrium two-dimensional melting is less understood than its equilibrium counterpart. Now it is shown that topologically driven melting in a two-dimensional crystal of charged colloids is the same irrespective of the mechanisms that generate the defects

Here the authors show that tissue-resident memory and exhausted T cells in tumors are distinct populations that are shaped by relative presence or absence of TCR signals, suggesting that a tailored therapeutic strategy is needed to target each subset.

The authors find that TDP-43 loss of function—the pathology defining the neurodegenerative conditions ALS and FTD—induces novel mRNA polyadenylation events, which have different effects, including an increase in RNA stability, leading to higher protein levels.

Whether high-order frontal lobe areas receive raw speech input in parallel with early speech areas in the temporal lobe is unclear. Here, the authors show that frontal lobe areas get fast low-level speech information in parallel with temporal lobe speech areas.

Sivanandan, Leitmann, and colleagues present the CellPaint-POSH platform, which combines pooled CRISPR screening with Cell Painting. Using self-supervised deep learning on cell images, the method enables discovery of gene function and biological networks.

Many ecological studies assume that space-for-time substitution approaches can be suitable proxies for unavailable time series. Here the authors show congruence between the two approaches in the direction but not the magnitude of grassland plant and arthropod community responses to land-use intensification.

Approximately 17% of meningiomas remain genomically uncharacterized. Here, the authors analyze 105 meningiomas without known driver mutations or significant copy number alterations and identify a subgroup of meningiomas, defined by FOS/FOSB gene fusions with distinctive transcriptomic and histopathological features.

Native state proteomics of PV interneurons revealed unique molecular features of high translational and metabolic activity, and enrichment of Alzheimer’s risk genes. Early amyloid pathology exerted unique effects on mitochondria, mTOR signaling and neurotransmission in PV neurons.

Chan and colleagues report that tumor ecosystem and microbiome features are associated with response to anti-PD-L1 avelumab plus chemoradiotherapy in patients with locally advanced head and neck cancer.

Humanity’s Last Exam, a multi-modal benchmark at the frontier of human knowledge, is designed to be an expert-level closed-ended academic benchmark with broad subject coverage.

iGluSnFR4f and iGluSnFR4s are the latest generation of genetically encoded glutamate sensors. They are advantageous for detecting rapid dynamics and large population activity, respectively, as demonstrated in a variety of applications in the mouse brain.

This study identifies key neurocognitive domains that distinguish patients with schizophrenia from healthy individuals using machine learning. Analyzing data from 1,304 participants, it demonstrates that verbal learning and emotion identification effectively classify conditions, promoting efficient neurocognitive profiling strategies.

Gas-phase actinium monofluoride (AcF) has been produced and spectroscopically studied at the CERN-ISOLDE radioactive ion beam facility; the results highlight the potential of ²²⁷AcF for exceptionally sensitive searches of CP violation.

In this study, the authors present an fMRI‑based signature of corticospinal connections, which predicts individual pain sensitivity, generalizes to patient cohorts, and tracks changes after brain stimulation, suggesting a biomarker to guide personalized pain care.

This work presents a deep learning model that predicts white matter tracts from anatomical MRI without diffusion data. It enables faster, accessible mapping of white-matter connections for research and clinical use.

SmartEM is a ‘smart’ pipeline for electron microscopy-based data acquisition for connectomics. In order to efficiently image large datasets, the approach involves imaging at short pixel dwell times and identifying problematic regions that are then imaged with longer dwell times and therefore higher quality.

STAARpipeline is a comprehensive framework for flexible and scalable rare-variant association analysis using whole-genome sequencing data and annotation information.

Investigating crystalline materials often requires calculations for many variations of a system, substantially increasing the computational burden. By training a transferable neural wavefunction across these variations, the cost can be reduced by approximately 50-fold for systems such as graphene and lithium hydride.

The factors governing mouse decision-making strategies in changing environments remain unclear. Here the authors use a 4-state input-driven model to show that past stimuli/choices drive switches in bias, while past rewards drive switches in engagement.

The Krakencoder can translate between structural and functional connectivity as well as between different atlases and processing choices via latent representations, as illustrated with datasets from the Human Connectome Project.

Electron tomography enables high-resolution 3D imaging but is limited by radiation sensitivity. Here, the authors demonstrate a phase retrieval method using multiple tilted HRTEM focal series, achieving 1.6 Å resolution with improved dose efficiency and elemental sensitivity.

scTranslator is a large-scale generative model that translates single-cell transcriptomic data to proteomic data and can assess the impact of gene perturbation on protein levels.

Gene regulation varies across cell types and states. Here, the authors present airqtl, a scalable method for single-cell eQTL mapping that enables efficient causal gene regulatory network inference, revealing the genetic drivers of cell state-specific regulation.

Microflora Danica—an atlas of Danish environmental microbiomes—reveals that although human-disturbed habitats have high alpha diversity, species reoccur, revealing hidden homogeneity.

In people with severe brain injuries, stimulation restored consciousness by engaging a deep brain circuit for wakefulness—revealing a target that may also guide treatment in stroke and epilepsy.

Here, the authors show learning tasks with similar structures can initially cause interference and slow down learning, but both the brain and artificial networks gradually reorganize information over time, enabling them to perform better and adapt more efficiently.

Amyloid plaques are a hallmark of Alzheimer’s disease. Better understanding of their biochemistry can inspire new biomarkers and therapeutics. Using multimodal mass spectrometry imaging, this work reveals surprising lipid heterogeneity in plaque microenvironments across the brain.

This Perspective examines the design of non-volatile memristive analogue in-memory computing tiles for deep neural network accelerators, considering the challenges and opportunities associated with designing the different components and providing projections for how technology scaling may impact key design elements.

Here, the authors examine the mechanisms behind cheatgrass’s successful invasion of North American ecosystems. Their genetic analyses and common garden experiments demonstrate that multiple introductions and migrations facilitated cheatgrass local adaptation.

Electrochemical acid-base production has attractive applications in mineral recovery and CO₂ removal, but current membrane-based designs are plagued by resistive losses. The authors report a membrane-less system generating useful acid and base solutions at high rates with less energy.

Water-vapor interfaces have been studied with many techniques, yet open questions persist about their electronic and molecular structure. Here, the authors demonstrate the application of soft x-ray second harmonic generation to study the water surface by leveraging attosecond pulses at the LCLS and a flat liquid sheet microjet, providing insights on the H-bond structure.

Fault tolerance is essential for reliable AI acceleration using novel memristive hardware. Yousuf et al. developed a training-free fault tolerance scheme and demonstrated on a 20,000-memristor prototyping platform that it outperforms other solutions.

The analysis of the energy spectrum of 36 million tritium β-decay electrons recorded in 259 measurement days within the last 40 eV below the endpoint challenges the Neutrino-4 claim.

Hunter et al. use RNA labelling to investigate RNA transfer between organs in mice. They show that RNA potentially moves en masse from liver to kidney and that this movement is augmented in acute liver injury, although the physiological relevance of the phenomenon is not yet known.

This study demonstrates that brain functional network imbalance appears linked to progressive brain atrophy and cognitive decline across the dementia spectrum.

By integrating a ribo-seq dataset of 195 individuals with RNA-seq and proteomics data from the same cohort, researchers found prevalent post-transcriptional attenuation of eQTL effects and identified omic-specific QTLs associated with schizophrenia.

Age-related macular degeneration (AMD) is a common cause of vision loss with a large genetic risk in older individuals. Here, for a high-risk AMD subtype, the authors identify an association with a chromosome 10 risk region containing a long non-coding RNA.

In this study Stubbs et al. find that neuroimaging neuroimaging abnormalities across substance use disorders map to a common brain network that is similar across imaging modalities and substance use categories.

Yashinskie, Zhu and colleagues show that p53 activation triggers increased synthesis and accumulation of phospholipids, with enhanced activation of autophagy and lysosomal catabolism programmes and increased reliance on lipid headgroup recycling.