Browse Articles

Autonomous biomedical research with an artificial intelligence agent

Biomedical research is increasingly constrained by repetitive, fragmented workflows that slow discovery. We introduce Biomni, a general-purpose biomedical artificial intelligence agent that autonomously executes diverse research tasks. To map the biomedical action space, Biomni’s action-discovery agent mines tools, databases, and protocols from thousands of publications across 25 domains, building a unified agentic environment. Its general-purpose architecture integrates large language model reasoning with retrieval-augmented planning and code-based execution, dynamically composing workflows without predefined templates. Systematic benchmarking shows strong generalization across heterogeneous tasks—causal gene prioritization, drug repurposing, rare-disease diagnosis, microbiome analysis, and molecular cloning—without task-specific tuning. Real-world case studies demonstrate Biomni interpreting multi-modal datasets, optimizing protein stability, orchestrating wet-lab instruments, and generating experimentally testable protocols. Biomni envisions artificial intelligence augmenting human scientists and accelerating discovery.

Colliding Forces — The Aging of the Baby Boom Generation and Contracting Nursing-Home Supply

AI-driven Chinese male facial skin phenotype profiling and its lifestyle correlation

The origin, history, and resistance architecture of an invasive urban malaria mosquito in Africa

The invasive urban malaria vector Anopheles stephensi threatens 126 million city dwellers in Africa. Controlling An. stephensi requires greater understanding of its origin, invasion dynamics, and insecticide resistance mechanisms. Analysis of 645 whole genomes sampled across Africa, the Middle East, and Asia supports an invasion scenario in which an initial South Asian introduction established a bridgehead population in Djibouti, which seeded distinct invasion fronts in Sudan, Ethiopia/Kenya, and Yemen. These incursions show contrasting rates and routes of spread shaped by landscape topology. Insecticide resistance is predominantly mediated by metabolic detoxification genes, with resistance haplotypes and copy-number amplifications introduced from South Asia. These findings, alongside a companion genomic resource, enable genomic surveillance of An. stephensi spread and resistance to aid control strategies.

Elevated glucose-to-platelet ratio predicts short- term and long-term mortality in critically ill patients with acute ischemic stroke

British ‘First Fleet’ brought smallpox to Australia

Colonists likely introduced the disease to a more populous continent than many imagined

Preserved structural and functional stability of high-dose aflibercept after compounding into prefilled syringes

AI systems devise hypotheses and ways to test them

Transient assembly of precision-tuned platinum-skin intermetallic catalysts for fuel cells

Highly efficient catalysts require precisely engineered intricate structures, yet conventional thermodynamically controlled syntheses often involve cumbersome procedures and limited structural precision. We report a nonequilibrium transient assembly strategy for the ultrafast synthesis of intricately structured nanocatalysts, including core-shell platinum (Pt)–skinned intermetallic nanocrystals exemplified by Pt@PtFe-i. By using a periodic thermal-pulse protocol to drive the continuous evolution of high-energy transient PtFe configurations, we achieved the synchronous assembly of a high-order PtFe intermetallic core and an atomic-layer-precise Pt skin. The Pt@PtFe-i catalyst exhibits coordination-dependent compressive strain within the Pt skin, creating a high density of highly active sites for the oxygen reduction reaction. The H 2 -air fuel cell with Pt@PtFe-i delivers a peak power of 1.25 watts per square centimeter at a cathode Pt loading of 0.1 milligrams per square centimeter, with a small peak power loss of 3.2% after 30,000 accelerated durability testing cycles.

An efficient and interpretable intrusion detection framework for software-defined networks with multi-class imbalanced data using genetic and GAN-based optimization

Antibodies sculpt adult brain circuits

Immune cells target hyperactive neurons to eliminate synaptic connections

ForNet: classification of critical forest acoustic events using discriminative CNN representations and ensemble learning

Abstract Various studies have been conducted in the field of sound event classification, with a specific emphasis on urban environments. However, there has been limited investigation in the realm of forest-oriented sound event classification. We introduce FSM5, a task-specific forest acoustic dataset curated from multiple publicly available sources to support forest sound event classification research, and propose a two-stage framework, ForNet, for effectively classifying critical sound events in forest. Initially, a Convolutional Neural Network (CNN) is used to extract discriminative audio embeddings. In the subsequent phase, ensemble classifiers such as XGBoost and Random Forest are employed for classification. The performance of MFCC, Log-Mel, and Mel spectrogram features is systematically evaluated, and the results indicate that MFCC and Log-Mel features significantly enhance classification performance. The findings indicate that combining handcrafted acoustic representations with CNN-derived embeddings yields improved performance compared to end-to-end CNN classification. The efficacy of deep CNN’s feature representation and the discriminative capability of shallow classifiers are evaluated. To showcase the reliability of ForNet, we have also conducted tests on the benchmark: Urbansound8k dataset. ForNet achieves an accuracy of 94% under 10-fold cross-validation on UrbanSound8K, outperforming several state-of-the-art methods, and attains an accuracy of 91.4% using 10-fold cross-validation on the FSM5 dataset.

USDA accelerates plan to close its flagship scientific campus

Agency says closing the research center will improve efficiency, but skeptics argue it will undercut research critical to farmers

Analysis and optimisation of the wastewater reinjection system at the Keshen gas field

A multi-agent system for automating scientific discovery

Ice as a geochemical reactor

Freezing redirects the fate of iron in Earth’s cryosphere

Performance of leading large language models in adhering to clinical guidelines for anaplastic thyroid cancer: a comparative study

Abstract Anaplastic thyroid cancer (ATC) is a rare, aggressive malignancy with poor prognosis. Adherence to guidelines from the National Comprehensive Cancer Network (NCCN), American Thyroid Association (ATA), and European Society for Medical Oncology (ESMO) is critical for optimal patient outcomes. As large language models (LLMs) increasingly enter clinical workflows, rigorous evaluation of their alignment with established guidelines is essential. We evaluated five leading LLMs for their ability to generate guideline-concordant responses to clinical questions about ATC. We conducted a comparative study in 2025 following TRIPOD-LLM (Transparent Reporting of a Multivariable Model for Individual Prognosis or Diagnosis, Large Language Models) guidelines. Seventy clinical questions of varying complexity were developed from ATA, NCCN, and ESMO guidelines. Three surgical oncology experts validated each question and subsequently evaluated responses from five LLMs: ChatGPT 4.1, ChatGPT 5, Gemini 2.5 Pro, Claude Sonnet 4, and DeepSeek R1. Each response was scored for relevancy, clarity, accuracy, and adequacy on a 5-point Likert scale. Inter-rater reliability was assessed using both intraclass correlation coefficients (ICC) and Gwet’s AC2 with ordinal weights. Model comparisons used the Kruskal-Wallis test with Dunn’s post-hoc analysis and Bonferroni correction. A pre-specified sensitivity analysis excluding the unblinded model (ChatGPT 5) was performed to confirm robustness. Significant performance differences emerged across all four metrics: accuracy ( p  = 0.007), adequacy ( p  = 0.003), clarity ( p  = 0.014), and relevance ( p  < 0.001). Gemini 2.5 Pro achieved the highest median accuracy (4.5), followed by DeepSeek R1 (4.4), while ChatGPT 4.1 scored lowest (4.0). ICC values ranged from 0.34 to 0.44 (poor to moderate), but Gwet’s AC2 yielded substantially higher estimates of 0.61 to 0.73 (moderate to substantial agreement), reflecting the impact of restricted score range on conventional reliability metrics. The sensitivity analysis excluding ChatGPT 5 confirmed the performance hierarchy among blinded models, with significance preserved or strengthened across all four metrics. Leading LLMs show variable capacity to align with ATC clinical guidelines. While top-performing models hold promise as supportive tools, their inconsistencies across domains and complexity levels preclude autonomous clinical use. These models should serve strictly as decision aids under expert supervision.

Does AI work for us, or do we work for AI? <b>The Reverse Centaur’s Guide to Life After AI</b> <i>Cory Doctorow</i> MCD, 2026. 240 pp.

There is little enthusiasm for a technology that turns human users into “reverse centaurs,” argues a critic

Identification of vacuum OLTC faults using improved multimodal PKO-SVM

Accelerating scientific discovery with Co-Scientist

Abstract Scientific discovery is driven by scientists generating hypotheses for complex problems that undergo rigorous experimental validation. To augment this process, we introduce Co-Scientist, a multi-agent artificial intelligence (AI) system built on Gemini for structured scientific thinking and hypothesis generation. Co-Scientist aims to help scientists discover new original knowledge. Conditioned on their research objectives and previous scientific evidence, it formulates demonstrably novel research hypotheses for experimental verification. The system’s design involves agents continuously generating, critiquing and refining hypotheses accelerated by scaling test-time compute. Key contributions include (1) a multi-agent architecture with an asynchronous task execution framework for flexible compute scaling, and (2) a tournament evolution process for self-improving hypotheses generation. Automated evaluations show continued benefits of test-time compute scaling, improving hypothesis quality over time. Although this is a general-purpose system, we focus the validation in three biomedical applications: drug repurposing; novel-target discovery 1 ; and explaining mechanisms of antimicrobial resistance 2 . Specifically, Co-Scientist helped to identify new drug-repurposing candidates and synergistic combination therapies for acute myeloid leukaemia that were validated through in vitro experiments. These real-world validations demonstrate the potential of Co-Scientist to accelerate scientific discovery and usher in an era of AI-empowered scientists.