Search

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.

An analog optical computer that combines analog electronics, three-dimensional optics, and an iterative architecture accelerates artificial intelligence inference and combinatorial optimization in a single platform, paving a promising path for faster and sustainable computing.

Sustained Mars exploration requires in situ synthesis of vital chemicals such as oxygen. Now a data-driven platform for synthesizing oxygen-producing electrocatalysts from Martian meteorites using robotics and artificial intelligence is developed, allowing automated screening of the optimal catalyst formula. This approach demonstrates materials discovery under challenging circumstances and without human intervention.

Diagnosis of invasive pulmonary nodules and prediction of tumor invasiveness in lung cancer patients remains challenging. Here, mass cytometry analysis of peripheral blood samples from 1032 patients and a machine learning approach, the SMART platform, allow differential diagnosis and surgical management of small pulmonary nodules.

Neural coupling is a challenge in understanding both brain function and advancing machine intelligence. Here, the authors introduce HoloBrain and HoloGraph, a brain-inspired framework that models oscillatory synchronization to overcome limitations of graph neural networks and enable more efficient, robust learning.

The plasma membrane Na⁺/H⁺ exchanger SOS1 is crucial for plant salt tolerance. Here the authors report the structure of SOS1 from Arabidopsis thaliana in two conformations, which provide structural and functional insight into SOS1 activity regulation.

CO electroreduction offers a sustainable route to acetate production but remains mechanistically unclear. Here, the authors develop an active learning multiscale framework revealing CH* binding strength as the key descriptor, predicting Cu/Pd and Cu/Ag as optimal catalysts.

An integrated surface-enhanced Raman scattering navigation instrument allows the intraoperative classification of malignant-grade prostate tumours by simultaneously assessing tissue acidity and prostate-specific antigen enzymatic activity.

Generation of mature oocytes requires tight regulation of a discontinuous meiotic cell cycle. Here they show that the acetyltransferase Nat10 mediates modification of RNAs targeted for degradation and find that this process is essential for female oocyte meiosis and maturation.

Collaborative augmented reconstruction (CAR) is a platform for large-scale reconstruction of neurons and other cells from multi-dimensional imaging datasets. It can be accessed from a variety of devices simultaneously for efficient and accurate reconstruction.

A new artificial intelligence model, DeepSeek-R1, is introduced, demonstrating that the reasoning abilities of large language models can be incentivized through pure reinforcement learning, removing the need for human-annotated demonstrations.

The authors identify 92 tick-associated viruses, including ten pathogenic to humans in northwest China and adjacent countries. They predict a significant increase of major pathogenic viruses and their vectors in high-risk regions.

This study used accelerometer data from participants of the UK Biobank to reveal associations between sedentary and physical activity levels and mortality in people with epilepsy. These associations differ between people with epilepsy and the general population.

Photon noise fundamentally limits 3D fluorescence imaging performance. Lu et al. proposed LF-denoising, a self-supervised transformer leveraging spatial-angular redundancy for denoising without temporal dependence, enabling high-fidelity intravital imaging.

Microwell-seq is used to generate single-cell, whole-body expression landscapes of zebrafish, Drosophila and earthworm. A deep-learning model, Nvwa, predicts gene expression from DNA sequence and identifies regulatory programs shared across eight species.

The spatiotemporal relationship between neuromodulator and neurotransmitter dynamics and neuronal activity remains unclear. Here, the authors performed dual-color imaging of calcium and ACh/5-HT signals across the fly brain, revealing a key role of ACh dynamics in stable odor representation.

The water electrolytic oxidation of graphite is a low-cost sustainable synthesis method for the large-scale production of graphene oxide (GO), but it is usually limited by non-uniform oxidation. Here, the authors report a liquid membrane electrolysis method enabling the high-yield synthesis of uniform monolayer GO at industrial scales.

The genome is actively organized in progenitors to regulate their capacity to produce different cell types. Here, the authors show that a nuclear architecture protein forms condensates and controls the physical location of the genomic locus of a key competence transcription factor.

It is generally more difficult to modify the buried interfaces compared to top interfaces in tin-lead perovskite subcells. Here, the authors employ a mercapto-functionalized mesoporous silica layer as a superstructure, realizing maximum efficiency of 29.6% for double-junction tandem solar cells.

Quantifying the function of enhancer sequences is crucial for systems biology. Here, the authors introduce an AI-added strategy to quantify the function of sequences surrounding TFBSs, which successfully modularizes the enhancer, and further achieves the re-design of enhancers with aimed functions.

Accurate prediction of peptidic hydrogels could prove useful for diverse biomedical applications. Here, the authors develop a “human-in-the-loop” approach that integrates coarse-grained molecular dynamics, machine learning, and experimentation to design natural peptide hydrogels.

A bioinspired mechanosensory skin integrating different neuromorphic processing modes enables robots to achieve highly efficient tactile texture discrimination.

Through development of non-fullerene acceptors, OPVs have reached efficiencies of 18%, yet the inadequate operational lifetime still poses a challenge for the commercialisation. Here, the authors investigate the origin of instability of NFA solar cells, and propose some strategies to mitigate this issue.

Using sequencing and haplotype-resolved assembly of 65 diverse human genomes, complex regions including the major histocompatibility complex and centromeres are analysed.

Identifying driver regulators in cell state transitions is key to decoding cellular function. Here, the authors present regX, an interpretable AI framework to prioritise potential driver TFs and cCREs from single-cell multiomics data, showing potential for understanding and manipulating cell states.

The topological properties of twisted bilayer MoTe₂ are thought to stem from a spatial texture in the layer polarization of the electronic wavefunctions. This polarization is now measured using scanning tunnelling microscopy.

Ammonia can be synthesized electrochemically from nitric oxide, but the catalytic performance has generally not been satisfactory. Here the authors report a highly active copper–tin alloy for nitric oxide reduction to ammonia, which they test in a flow cell and a membrane electrode assembly.

Hunting behavior typically contains a sequential motor program, including search, chase, attack, and consumption. Here, the authors show that periaqueductal gray neuronal ensembles encode the sequential hunting motor program, which might provide a framework for decoding complex instinctive behaviors.

The FarmGTEx Project aims to understand genetic control of gene activity under diverse biological and environmental contexts in domestic animals, providing a foundation for improving animal precision breeding, adaptation and human health.

Liu, Gu, Chen, Zhang et al. propose KUTS, a foundational triage model that integrates knowledge prompt-tuning and uncertainty-based classification to enhance emergency acuity level prediction. The model achieves superior AUC performance across all levels, especially for moderate acuity cases, offering a robust and efficient solution for real-world triage applications.

Identifying interventions that can induce a desired effect is challenging owing to the combinatorial number of possible choices in design space. Zhang and colleagues propose an active learning approach with theoretical guarantees to discover optimal interventions in causal models, and demonstrate the framework in the context of genetic perturbation design using single-cell transcriptomic data.