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.

AI is a powerful tool for facilitating data science programming in biomedicine. This study benchmarks the performance of existing tools and introduces a reliable AI agent copilot.

Using generative AI, Insilico Medicine developed an oral PROTAC that potently inhibits and degrades PKMYT1, a synthetically lethal target in cancer, demonstrating high selectivity and strong antitumor efficacy in preclinical models.

Preferred orientation in cryo-EM often limits structural resolution. Here, the authors present cryoPROS, a novel computational method that co-refines raw and deep generative auxiliary particles to correct misalignment, enabling near-atomic resolution from challenging datasets.

An annotation-free vision–language model based on contrastive learning performs pathology localization and clinical diagnosis on a wide variety of medical image datasets.

Fibre integrated circuits created by first preparing circuitry on extremely thin and flexible substrates and then mechanically rolling them up demonstrate microdevice density reaching 100,000 transistors per centimetre and excellent stability under harsh service conditions.

By analyzing the structures of the Polycomb repressive deubiquitinase complex, which catalyzes the deubiquitination of H2AK119Ub on nucleosomes, Zhang, Zheng, Tong and Deng et al. reveal that ubiquitin acts as a proteinaceous glue to fasten the intersubunit interaction within the complex to allosterically enhance its activity.

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.

Xiao et al. introduce ‘capability density’, defined as capability per parameter, as a metric for evaluating large language models. They report an empirical trend, the ‘densing law’, which states that capability density doubles approximately every 3.5 months, indicating that equivalent model performance can be achieved with exponentially fewer parameters over time.

FlyWire presents a neuronal wiring diagram of the whole fly brain with annotations for cell types, classes, nerves, hemilineages and predicted neurotransmitters, with data products and an open ecosystem to facilitate exploration and browsing.

Understanding the genetic factors that regulate panicle architecture is important for improving rice yields. In this study, Li et al. demonstrate that the gene PTB1is critical for regulating panicle seed setting rate and pollen tube growth.

Artificial intelligence (AI)-based docking and scoring methods demonstrate considerable potential for virtual drug screening. Gu et al. go further by assessing the structural rationality of AI-predicted complex conformations from various sources.

BioPathNet predicts drug–disease indications, synthetic lethality and lncRNA–mRNA interactions using a deep learning framework.

The integration of organic and enzymatic synthesis enhances molecule construction efficiency. Here, the authors present ChemEnzyRetroPlanner, an AI-driven platform for automated hybrid synthesis planning, improving synthesis route efficiency and sustainability.

DeepRETStroke is a deep learning system using retinal photographs to detect silent brain infarction and predict future stroke events.

In this Perspective, members of the Aging Biomarker Consortium outline the X-Age Project, an Aging Biomarker Consortium plan for building standardized aging clocks in China. The authors discuss the project roadmap and its aims of decoding aging heterogeneity, detecting accelerated aging early and evaluating geroprotective interventions.

Trained and validated on multimodal data from 14.5 million images from multicountry datasets, a foundation model is shown to increase diagnostic and referral accuracy of clinicians when used as an assistant in a trial involving 16 ophthalmologists and 668 patients.

Large language models (LLMs) can be useful tools for science, but they often lack expert understanding of complex domains that they were not trained on. Zhang and colleagues fine-tuned a LLaMA-2-7b-based LLM with questions on organic chemistry reactions.

Liu et al. report Chinese normative lifespan brain charts showing later neurodevelopmental milestones than those detected in Western cohorts. Individual deviations from these norms are valuable in assessing clinical risk and outcomes.

Two potent mitophagy inducers, identified and characterized via unsupervised machine learning and a cross-species screening approach, ameliorated the pathology of Alzheimer’s disease in worms and mice.

Pertpy serves as a one-stop and performant framework for single-cell perturbation data analysis.

Computer-aided synthesis-planning methods have significantly assisted synthesis planning. In this work, the authors present RSGPT, a generative model pre-trained on ten billion data points, achieving state-of-the-art performance for synthesis planning

A deep learning algorithm shows promising performance in predicting progression to diabetic retinopathy in patients, up to 5 years in advance, potentially providing support for medical treatment decisions and indications for personalized screening frequency in a real-world cohort.

An emerging research area in AI is developing multi-agent capabilities with collections of interacting AI systems. Andrea Soltoggio and colleagues develop a vision for combining such approaches with current edge computing technology and lifelong learning advances. The envisioned network of AI agents could quickly learn new tasks in open-ended applications, with individual AI agents independently learning and contributing to and benefiting from collective knowledge.

The development of machine learning systems has to ensure their robustness and reliability. The authors introduce a framework that defines a principled process of machine learning system formation, from research to production, for various domains and data scenarios.

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 ratio of tumour area to metastatic lymph node area (T/MLN) is a clinical indicator that can improve prognosis prediction of gastric cancer. Here, the authors use machine learning on whole slide images to generate a method that can predict metastatic lymph nodes and obtain T/MLN.

Atrial fibrillation is a prevalent and serious arrhythmia associated with increased morbidity and mortality worldwide. Here the authors show a contactless, operation-free, and device-free AF detection framework utilizing artificial intelligence-powered radio technology, achieving performance comparable to conventional ECG-based methods.

MnBi₂Te₄ has an appealing combination of topological bands and magnetic ordering. While chemical doping with Sb can be used to tune these properties, it typically comes with an increase in defect density. Here, Chen, Wang, Li, Duan, and coauthors demonstrate a defect engineering approach that preserves the topological and magnetic properties of Mn(Bi_1-xSb_x)₂Te₄.

A novel covalent inhibitor, ISM3312, targets the main protease of multiple human coronaviruses, including drug-resistant strains, and shows broad antiviral activity. It offers a promising therapeutic strategy against current and future coronavirus threats.

SIDISH integrates single-cell and bulk RNA sequencing data using deep learning to identify high-risk cell populations and prognostic biomarkers, enabling in silico perturbations that could guide precision therapeutics and advance personalized medicine.

Mendelian randomization is used to evaluate genetic associations between multi-organ imaging traits and clinical outcomes, showing genetic links between organ systems in a variety of diseases.

The impact of fluorine doped tin oxide (FTO) on device longevity has been overlooked. Here, the authors employ thermal evaporated yttrium oxide to strengthen structural stability of FTO, achieving maximum efficiency of 26.48% and 28.47% in regular and all-perovskite tandem structures, respectively.

Simultaneously capturing polarization and frequency is challenging. This work develops a metasurface-based intelligent photodetector, which can detect arbitrary polarization and frequency within 0.3-1.1 THz with a total error smaller than 5.1%.

Nitric oxide in known to have a protective effect in cerebrovascular disease. Here, the authors report on MoS₂ an eNOS mimetic which releases NO while avoiding toxic ONOO^- production, demonstrating therapeutic effects in inhibiting several damage pathways in cerebral ischemia-reperfusion injury.

Here the authors report NiGa₂O_4–x(OH)_y for light-driven CO₂ hydrogenation to methanol. The surface Lewis acid–base pairs and -OH groups act as conduits for H^- /H⁺ transport to active sites, enhancing photocatalytic methanol production.

Tailored to provide diabetes management recommendations from large training and validation datasets, an artificial intelligence system integrating language and computer vision capabilities is shown to improve self-management of patients in a prospective implementation study.

Scattering in the archetypal oxide SrRuO₃ is shown to enhance orbital currents. This counter-intuitive effect establishes a transformative paradigm for energy-efficient spintronic devices.

Humans strive to design safe AI systems that align with our goals and remain under our control. However, as AI capabilities advance, we face a new challenge: the emergence of deeper, more persistent relationships between humans and AI systems. We explore how increasingly capable AI agents may generate the perception of deeper relationships with users, especially as AI becomes more personalised and agentic. This shift, from transactional interaction to ongoing sustained social engagement with AI, necessitates a new focus on socioaffective alignment—how an AI system behaves within the social and psychological ecosystem co-created with its user, where preferences and perceptions evolve through mutual influence. Addressing these dynamics involves resolving key intrapersonal dilemmas, including balancing immediate versus long-term well-being, protecting autonomy, and managing AI companionship alongside the desire to preserve human social bonds. By framing these challenges through a notion of basic psychological needs, we seek AI systems that support, rather than exploit, our fundamental nature as social and emotional beings.

A consensus cell type atlas for the fly brain provides both an intellectual framework and open-source toolchains for brain-scale comparative connectomics.

GenomePAM compares activities and fidelities among different Cas nucleases on thousands of match and mismatch sites across the genome using a single gRNA.