The creation of effective governance mechanisms for artificial intelligence (AI) agents requires a deeper understanding of their core properties and the implications they have for deployment. This paper provides a characterization of AI agents that focuses on four dimensions: autonomy, efficacy, goal complexity and generality. We propose different gradations for each dimension and argue that each dimension raises unique questions about the design, operation and governance of these systems. Moreover, we draw on this framework to construct 'agentic profiles' for different kinds of AI agent. These profiles help to illuminate cross-cutting technical and non-technical governance challenges posed by different classes of AI agents, ranging from narrow task-specific assistants to highly autonomous general-purpose systems. By mapping out key axes of variation and continuity across four dimensions, agentic profiles provide developers, policymakers and members of the public with guidance for effective AI governance.

Defence against viral infection is a conserved feature of all cellular life. From single-cell bacteria to humans and complex multicellular animals, constitutive and inducible forms of immunity are required to inhibit viruses and safeguard cellular fitness. Recent studies reveal that the components of human antiviral immunity are surprisingly ancient, originating billions of years ago in bacteria as pathways that defend against phage replication. The unification of previously disparate fields of human and bacterial immunity creates a foundation to explain key features of host-virus interactions. This Review defines principles of pathogen recognition, signal amplification and immune effector function that shape mechanisms of antiviral immunity that are shared across kingdoms of life. Shared forms of immunity, including cGAS-STING, inflammasomes, argonautes and viperin, reveal ancient features of antiviral defence. Similarly, direct comparisons of pattern recognition receptors and interferon-stimulated genes in human cells with CRISPR immunity and anti-phage defence systems in bacteria explain prevalent strategies to effectively sense and inhibit viral replication. Cross-kingdom analysis reveals universal rules that control host-virus interactions and highlights open questions in understanding of antiviral immunity.

Urban environments have long been a central focus for crime researchers across diverse disciplines. Over the past few decades, this heterogeneous area of inquiry has experienced substantial methodological and empirical change, driven by the emergence of novel datasets and the increasing use of flexible computational methods. In this Review, we take stock of this evolution and examine the potential that these developments hold for advancing urban crime research, while also addressing the persistent challenges that continue to shape the field. Building on this overview, we emphasize the promise that computational methods offer for more rigorous causal inference beyond traditional prediction tasks. Finally, we outline three key directions for future research to ensure that new data and computational tools are used effectively: greater integration across disciplines, improved open science standards and a broader scope of inquiry beyond Western contexts. In doing so, we aim to support more rigorous research and inform the development of more effective policies for safer and more sustainable cities worldwide.

To advance the computational simulation of cellular life, we propose a virtual yeast, an artificial intelligence (AI)-driven agent that models eukaryotic cellular behaviours by integrating multimodal biological data, mechanistic reasoning and active experimentation using Saccharomyces cerevisiae as a genetically tractable and data-rich model system. Cellular complexity is decomposed into eight function-centred modules, spanning genetic, metabolic and structural systems, each realized as a domain-specific AI tool coordinated through a large language model-based orchestration layer. Built on three data pillars, namely, mechanistic knowledge, subcellular architecture and dynamic states, the system integrates representation learning and generative modelling within a closed-loop learning pipeline that autonomously designs and executes experiments. The virtual yeast serves as both a conceptual and an operational platform to optimize biosynthetic pathways, support the generation and prioritization of hypotheses across diverse cellular processes, and accelerate target discovery. By coupling biological realism with autonomous AI reasoning, the virtual yeast establishes a generalizable blueprint for constructing virtual eukaryotic cells and advancing synthetic biology.

Nucleoside-modified mRNA-lipid-nanoparticle (mRNA-LNP) vaccines confer a high level of protection against severe COVID-19 and, since their first authorization for human use in 2020, have saved millions of lives. The efficacy of this vaccine platform relies on the induction of powerful and coordinated innate and adaptive immune responses. A deep understanding of the mechanisms of action by which mRNA-LNP vaccines drive protective immunity is crucial for advancing the development of next-generation mRNA vaccines with improved immunogenicity and tolerability. A flurry of recent studies has shed light on aspects of this vaccine modality's modus operandi. Nonetheless, key gaps in knowledge remain, including understanding how LNPs are sensed by the immune system and exert their adjuvant activity, identifying the specific signals and cellular pathways critical for eliciting protective immune responses and determining whether it is feasible to uncouple vaccine immunogenicity and reactogenicity. Here we review the known and unknown features of the immunological mechanisms of mRNA-LNP vaccines for infectious diseases. Furthermore, we discuss how the components of this vaccine platform can be modified to fine-tune immune responses against challenging pathogens for which effective vaccines do not exist or need improvement.

At present, more than 700 million people live with caloric hunger, and more than two billion suffer from micronutrient deficiencies, known as 'hidden hunger'. From an agricultural viewpoint, three major objectives need to be worked towards simultaneously to achieve zero hunger (the United Nations Sustainable Development Goal 2): (1) enhanced yield; (2) higher vitamin and mineral density to sustain recommended daily intake (multi-biofortification); and (3) enhanced climate-change resilience. Although the Green Revolution increased global calorie production, it exacerbated hidden hunger by prioritizing high yield over nutritional quality. Stress from global climate change has been shown to reduce the densities of several micronutrients. CRISPR-Cas, which allows genome editing with extremely high precision, has emerged as a groundbreaking breeding technology that has already been adopted by many countries. Here we examine how CRISPR-Cas-based approaches could be used to achieve biofortification targets by enhancing micronutrient densities to the levels necessary to alleviate dietary vitamin and mineral deficiencies. Given the limited time frame available to achieve zero hunger, we argue that CRISPR-Cas technologies should be combined with metabolic engineering based on transformation and other technologies. We also consider untapped resources beyond metabolic pathways and current CRISPR-Cas methodologies to address one of the most important societal issues of the twenty-first century.

Despite large uncertainties associated with future mass loss from the Antarctic Ice Sheet, ice-sheet models show that the rate of sea-level rise from Antarctic ice loss in 2025 is strongly predictive of the rate for the next several decades, regardless of emission pathway or model complexity. This finding is robust across all models that were considered in the Intergovernmental Panel on Climate Change Sixth Assessment Report global mean sea-level projections, including the low-likelihood, high-impact scenarios of sea-level rise. Given this strong near-term decadal predictability, ice-sheet models that can accurately reproduce present-day ice-mass loss provide a reliable basis for near-term sea-level planning and adaptation through to mid-century. The predictability breaks down by the end of the twenty-first century as feedbacks, such as those related to marine ice-sheet retreat, begin to emerge, leading to accelerating ice loss. Drawing on these results, we identify key feedback mechanisms that can account for the transition between near-term decadal predictability and the longer-term, feedback-driven evolution, and suggest priorities for ice-sheet model development aimed at resolving long-term sea-level rise uncertainty.

Interpreting the association of genetic variants with complex traits can be improved by gaining a greater understanding of the molecular consequences of these variants. Although genome-wide association studies (GWAS) for complex diseases routinely profile over one million individuals1-5, studies of molecular traits have lagged behind. Here we performed a GWAS meta-analysis for 249 circulating metabolic traits in the Estonian Biobank and the UK Biobank in up to 619,372 individuals. We identified 88,127 common and low-frequency locus-trait associations from 8,398 loci that converged on shared genes and pathways. Using statistical fine mapping, systematic phenome-wide colocalization and cis-Mendelian randomization, we explored putative causal links between metabolic traits and disease outcomes. We predict that although plasma branched-chain amino acids (BCAAs) have been associated with type 2 diabetes in observational studies6,7, lowering BCAA levels by targeting the BCAA catabolism pathway is unlikely to reduce type 2 diabetes risk. Leveraging our large sample size and high-quality genotype imputation, we found that 19.4% of the confidently fine-mapped variants had minor allele frequencies between 0.1 and 1%, and these variants were twofold enriched for predicted missense and splice-altering variants. Our results highlight the value of integrating low-frequency variants into genetic association studies.

Two decades ago, two papers in Nature described how PARP inhibitors selectively killed cells deficient in the BRCA1 or BRCA2 tumour suppressor genes, observations that led to the first clinically approved treatment of a cancer with a targeted therapy selected based on a germline biomarker. This work was recognized by Nature as one of the top 20 discoveries in cancer in the twenty-first century and provides a compelling example of leveraging fundamental biology discovery for patient benefit. For people with specific forms of breast, ovarian, prostate or pancreatic cancer, these discoveries changed their care, enabling the use of more effective and better tolerated targeted therapies that both improve survival and quality of life. This in turn extended the role of germline BRCA1 and BRCA2 mutation testing from determining risk in the unaffected, to being a companion diagnostic biomarker used to determine therapy for a patient with cancer. The significance of these discoveries spread beyond BRCA1-mutant and BRCA2-mutant cancers: the synthetic lethal concept of the BRCA-PARP inhibitor effect highlighted the myriad levels of functional redundancy that exist in tumour cells and stimulated the search for other tumour-specific synthetic lethal effects that could be exploited therapeutically. Here we distill the learnings from the past two decades in this field.

With deep-learning-powered advances in protein design methods, there is an ongoing paradigm shift in protein engineering from random selection to intentional computational design methods. Here we describe the current state of de novo protein design. While there is still room for improvement in success rates and activities, the long-standing challenges of designing new protein structures, assemblies and protein binders are close to being solved. The key current questions in these areas are not how to design, but what to design, and open-source design methodology such as RFdiffusion and ProteinMPNN together with protein structure prediction tools enable biochemists and molecular biologists to broadly explore possible applications. There has also been considerable progress in the de novo design of small-molecule target binders, enzymes and multistate protein systems. Current challenges for methods development include design of catalysts for reactions with high energy barriers and, more generally, design of switches and nanomachines that integrate binding, conformational change and catalysis. Over the next five to ten years, we anticipate the design of sophisticated protein nanomachines and materials with functionality ranging far beyond that generated during natural evolution for a wide range of applications in medicine, technology and sustainability.

Almost all tumours carry one or more cancer driver mutations, which are essential for cell transformation. However, recent advances in cancer genomics have demonstrated that normal human tissues contain millions of cells carrying known driver mutations, while preserving homeostasis. Most of these mutated cells will never transform into tumours. Moreover, studies of known or suspected human carcinogens have shown that the majority are not mutagens. These observations suggest that exogenous carcinogenic exposures might increase cancer risk by modifying selective constraints, promoting the expansion of pre-existing clones carrying specific oncogenic mutations. In this Review, we propose a synthesis between ideas put forward almost a century ago based on seminal experiments on carcinogen-induced tumours in mice, observations made by cancer epidemiologists over several decades, and the recent revelation that normal human tissues are a patchwork of mutant clones. The repeated interplay between variation and selection-the first principles of Darwinian evolution-underlies the clonal selection leading to tumorigenesis. A deeper understanding of these processes can enhance prospects for cancer prevention by eliminating or mitigating the effects of environmental or endogenous tumour promoters.

The ability of cancer cells to consistently escape therapy highlights their remarkable adaptive potential. A longstanding debate in cancer research concerns whether drug resistance originates primarily from mutational processes or through cellular plasticity. Emerging evidence has suggested that adaptive cellular states arise through phenotypic plasticity triggered by intracellular stress signals. Here we propose a theoretical framework for how such cellular adaptation in cancer drug resistance could be 'learned' by the AP-1 family of transcription factors. We highlight key AP-1 properties, including regulatory combinatorics, stress-induced feedback and cellular memory, and argue that this system constitutes a molecular framework for establishing drug-resistant cellular states. Finally, we discuss the potentially broad relevance of this adaptation mechanism beyond cancer.

Metabolomics has matured into a powerful approach for probing metabolism, offering readouts that closely reflect cellular and organismal function in health and disease. Here we highlight two rapidly advancing frontiers: single-cell metabolomics and population-scale metabolomics. Single-cell metabolomics resolves the metabolic states of individual cells, uncovering cell-to-cell heterogeneity and spatial organization within tissues. Population-scale profiling profiles metabolites across large cohorts, enabling the discovery of markers of disease, environmental exposures and genetic variation. Although these approaches operate at different scales, they face shared challenges-including metabolite identification, quantification and multimodal data integration-and offer common advantages, such as the ability to capture non-genetic influences on phenotype and to scale to high throughput. We propose that continued advances in scalability will bring these domains together, enabling the construction of comprehensive metabolic atlases that chart cellular and interindividual variation and provide training data for foundation models of metabolism. By integrating cellular and population-level insights, single-cell and population-scale metabolomics promise to advance our understanding of metabolism across biology, medicine and pharmacology.

Advances at the intersection of biotechnology, artificial intelligence, electronics and materials science are reshaping how drugs can be delivered inside the body. Intelligent and miniaturized drug delivery devices (IMDDDs) leverage these technologies to achieve precise pharmacokinetics, targeted distribution and programmable release while minimizing toxicity and improving patient adherence. Unlike conventional approaches, IMDDDs can incorporate real-time sensing and adaptive control, enabling drug administration that is more precise and more responsive to dynamic physiological conditions. In this Review, we outline key categories and design principles, highlight artificial intelligence technologies for augmenting performance, discuss potential clinical applications across cancer, diabetes, cardiovascular disease, vaccination and beyond, and examine translation challenges and opportunities. By uniting engineering innovation with medical need, IMDDDs exemplify the next generation of drug delivery technologies.

The impacts of sea-level rise and other hazards on the coasts of the world are determined by coastal sea-level height and land elevation1. Correct integration of both aspects is fundamental for reliable sea-level rise and coastal hazard impact assessments2,3, but is often not carefully considered or properly performed. Here we show that more than 99% of the evaluated impact assessments handled sea-level and land elevation data inadequately, thereby misjudging sea level relative to coastal elevation. Based on our literature evaluation, 90% of the hazard assessments assume coastal sea levels based on geoid models, rather than using actual sea-level measurements. Our meta-analyses on global scale show that measured coastal sea level is higher than assumed in most hazard assessments (mean offsets [standard deviation] of 0.27 m [0.76 m] and 0.24 m [0.52 m] for two commonly-used geoids). Regionally, predominantly in the Global South, measured mean sea level can be more than 1 m above global geoids, with the largest differences in the Indo-Pacific. Compared with geoid-based assumptions of coastal sea level, the measured values suggest that with a hypothetical 1 m of relative sea-level rise, 31-37% more land and 48-68% more people (increasing estimates to 77-132 million) would fall below sea level. Our results highlight the need for re-evaluation of existing coastal impact assessments and improvement of research community standards, with possible implications for policymakers, climate finance and coastal adaptation.

The question of whether large language models (LLMs) can exhibit moral capabilities is of growing interest and urgency, as these systems are deployed in sensitive roles such as companionship and medical advising, and will increasingly be tasked with making decisions and taking actions on behalf of humans. These trends require moving beyond evaluating for mere moral performance, the ability to produce morally appropriate outputs, to evaluating for moral competence, the ability to produce morally appropriate outputs based on morally relevant considerations. Assessing moral competence is critical for predicting future model behaviour, establishing appropriate public trust and justifying moral attributions. However, both the unique architectures of LLMs and the complexity of morality itself introduce fundamental challenges. Here we identify three such challenges: the facsimile problem, whereby models may imitate reasoning without genuine understanding; moral multidimensionality, whereby moral decisions are influenced by a range of context-sensitive relevant moral and non-moral considerations; and moral pluralism, which demands a new standard for globally deployed artificial intelligence. We provide a roadmap for tackling these challenges, advocating for a suite of adversarial and confirmatory evaluations that will enable us to work towards a more scientifically grounded understanding and, in turn, a more responsible attribution of moral competence to LLMs.

Biofluid-based biomarkers have transformed neurodegenerative disease research and care, providing insights into the molecular underpinnings of Alzheimer's disease (AD) and other neurodegenerative dementias. This Review provides an update on recent developments in biofluid-based biomarkers for amyloid-β (Aβ) pathology, tau pathology, neurodegeneration, glial reactivity, α-synuclein pathology, TAR DNA-binding protein 43 (TDP-43) pathology, synaptic pathophysiology and cerebrovascular disease-pathologies and processes that are all relevant to neurodegenerative dementias. Complementing longstanding cerebrospinal assays, improved technologies now facilitate the detection of molecules linked to neurodegenerative brain changes at very low concentrations in the blood. This promises to complement the clinical evaluation of suspected neurodegenerative disease in healthcare with molecular phenotyping biomarkers that will help to link the clinical symptoms to ongoing pathophysiological processes in the brain and improve how patients are referred to specialty clinics for initiation and monitoring of molecularly targeted treatments. Clinically relevant breakthroughs such as the use of anti-Aβ monoclonal antibodies to address Aβ pathology in AD serve as important proof-of-concept examples of how the field is advancing toward molecularly informed prevention and treatment. This Review provides an overview of the most established biofluid-based biomarkers currently in use and offers practical guidance on their interpretation and implementation in clinical settings.

The recent discovery of altermagnetism was in part motivated by the research of compensated magnets towards highly scalable spintronic technologies. Simultaneously, altermagnetism shares the anisotropic higher-partial-wave nature of ordering with unconventional superfluid phases, which have been at the forefront of research for the past several decades. These examples illustrate the interest in altermagnetism from a broad range of science and technology perspectives. Here we review the symmetry, microscopy and spectroscopy signatures of altermagnetism. We describe the spontaneously broken and retained symmetries that delineate altermagnetism as a distinct phase of matter with d-, g- or i-wave compensated collinear spin ordering. In materials ranging from weakly interacting metals to strongly correlated insulators, the microscopic crystal-structure realizations of the altermagnetic symmetries feature a characteristic ferroic order of anisotropic higher-partial-wave components of atomic-scale spin densities. These symmetry and microscopy signatures of altermagnetism are directly reflected in spin-dependent electronic spectra and responses. We review salient band-structure features originating from the altermagnetic ordering, and from its interplay with spin-orbit coupling and topological phenomena. Throughout, we compare altermagnetism with traditional ferromagnetism and Néel antiferromagnetism, and with magnetic phases with symmetry-protected compensated non-collinear spin orders. We accompany the theoretical discussions with references to relevant experiments.

African populations remain substantially under-represented in research studies and global genomic databases. As the ancestral home of anatomically modern humans, Africa holds pride of place regarding human genetic diversity, with a deep and complex evolution over hundreds of thousands of years of human migration, admixture, and exposure to climate changes and infectious agents. Yet our present view of genomic diversity in Africa is sparse and poorly captures the rich variation across its more than 2,000 ethnolinguistic groups. To enhance representation, the Assessing Genetic Diversity in Africa (AGenDA) project, under the umbrella of the Human Heredity and Health in Africa (H3Africa) consortium, identified under-represented groups across nine different African countries for human whole-genome sequencing, with a view to enriching global datasets. Here we share our processes, including community engagement, obtaining ethics approvals, navigating legal compliance and developing a common governance framework. AGenDA is a testament to the determination of the scientific community to undertake research in challenging environments. It is led from Africa by African investigators who are the decision-makers in data-sharing processes. AGenDA is a step towards greater African representation in global genomic datasets to advance genomic research towards enabling precision medicine for Africa and the world.

In 2020, a global surge of activism linked to the Black Lives Matter movement prompted scientists to stage an academic 'strike', drawing attention to the ethical responsibility of addressing systemic racism. This catalysed debate in conservation, adding urgency to decades of scholarship on marginalization. In this Perspective, we review this literature and examine how exclusion in conservation persists across intersections of race, class, urban-rural divides, nationality and power dynamics from local to global levels. We highlight how marginalization and 'othering' disproportionately affect Black, Indigenous and people of colour (BIPOC) communities, especially in the Global South. Expansion of protected areas and the prioritization of individual animal lives over human well-being can intensify such inequities. We propose a framework for more inclusive conservation: recognizing and supporting human rights, ensuring local community agency, challenging entrenched norms in BIPOC engagement, and fostering educational opportunities led by and for BIPOC communities. Amid shifting global politics, including reduced US federal support for social and conservation issues, this framework provides guidance to counter racism and exclusion. By rethinking conservation practice, it seeks to build long-lasting, equitable and inclusive approaches that respect both people and nature.