Search

BioMedAgent is a self-evolving LLM multi-agent framework that learns to use various bioinformatics tools and chain them into executable workflows for autonomously carrying out diverse biomedical data tasks initiated by natural-language prompts.

Here, the authors introduce information isotopes, a novel framework for tracing unauthorized training data in black-box AI systems, offering a practical tool for auditing data provenance and protecting data rights.

Generative large language models (for example, ChatGPT and Claude) were used to score Big-Five traits from open-ended thoughts and video diaries. Large language model scores aligned with self-reports and predicted behaviour, showing personality’s expression in everyday language.

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.

The Fair Human-Centric Image Benchmark (FHIBE, pronounced ‘Feebee’)—an image dataset that implements best practices for consent, privacy, compensation, safety, diversity and utility—can be used responsibly as a fairness evaluation dataset for many human-centric computer vision applications.

While there is a great potential to use generative AI to address socio-economic challenges, there are also obstacles for creating locally adapted AI tools for fair development in LMICs, which are all discussed in this Perspective.

Developing efficient battery electrolytes is a slow and complex challenge. Here, authors create a deep active learning framework that rapidly identifies high-performance materials and transfers knowledge to new designs, tripling battery lifespan while requiring significantly fewer experiments.

MedHELM, an extensible evaluation framework including a new taxonomy for classifying medical tasks and a benchmark of many datasets across these categories, enables the evaluation of large language models on real-world clinical tasks.

Protein engineering is often hindered by the gap between modelling and experiment. Here, authors present ORI, a closed-loop computational framework to iteratively optimize proteins, achieving 100-fold higher activity, enhanced thermal stability, and the design of multifunctional enzymes.

Systemic lupus erythematosus (SLE) is a complex autoimmune condition and treatment options have limited efficacy. Here, by analysis clinical samples and employing preclinical models, the authors show that cGAMP levels are elevated in the plasma of patients with SLE and that loss of cGAS ameliorates symptoms in an imiquimod-induced lupus model. Importantly, they identify prasugel as a cGAS inhibitor with promising therapeutic potential.

This study identifies shared and distinct brain patterns underlying reappraisal and acceptance, showing how emotion regulation operates in dynamic naturalistic contexts and highlighting potential clinical markers.

The vision of human-level machine intelligence laid out by Alan Turing in the 1950s is now a reality. Eyes unclouded by dread or hype will help us to prepare for what comes next.

This study uses two large fundus cohorts to show that retinal foundation models generalize well across populations but exhibit age-related fairness gaps. It highlights how pre-training data composition shapes model generalisability and fairness.

Detailed oxygen maps of spiral galaxy NGC 1365 and cosmological simulations trace when its disk, bar and outer gas formed, showing that chemical fingerprints can reveal the merger and growth history of a galaxy.

ERAST speeds up homology search and provides a vector database for 1 billion biological sequences.

The ring-opening functionalization of strained rings represents a highly valuable transformation, but existing strategies predominantly focus on ring-opening difunctionalization. Now a multisite, programmable, divergent ring-opening functionalization of strained rings via the electro-oxidation of continuous C–C bond and multiple C(sp³)–H bonds has been developed.

Multimodal AI combines different types of data to improve decision-making in fields such as healthcare and engineering, but work so far has focused on vision and language models. To make these systems more usable in the real world, Liu et al. discuss the need to develop approaches with deployment in mind from the start, working closely with experts across relevant disciplines.

Ten of Nature’s recent contributors share their current book obsessions.

A new AI framework autonomously explores single-cell RNA sequencing datasets and produces creative, biologically relevant findings.

While leading tech companies race to build ever-larger models, researchers in Brazil, India and Africa are using clever tricks to remix big labs’ LLMs to bring AI to billions of users.

First-of-its-kind US bill would address the environmental costs of the technology, but there’s a long way to go.

Synchronous activation of climbing fibres engages disinhibitory circuitry to promote large dendritic calcium signals in Purkinje cells that are necessary to promote cerebellar learning.

AIaided diagnosis is an exciting area of cancer research, however, large scale training is limited by the availability of imaging datasets. Here, the authors develop FreeTumor as a generative AI framework to develop realistic tumor images for clinical application.

A secure artificial intelligence framework is introduced that leverages homomorphic encryption to safeguard sensitive information in deep reinforcement learning, achieving accurate decision-making and ensuring data privacy and confidentiality.

Ying and colleagues present PhyE2E, an AI framework incorporating symbolic search techniques for discovering physics formulas directly from data. The method has already led to improvements in space physics models when compared, for example, with NASA’s 1993 formula for solar activity.

While deep learning has enabled the decoding of language from intracranial brain recordings, achieving this with non-invasive recordings remains an open challenge. Here the authors introduce a deep learning pipeline to decode individual words from EEG and MEG signals

Including data for more than 125,000 compounds and biological agents, a dataset leverages a variety of sources including trial registries, news and stock prices and artificial intelligence to annotate outcomes and features automatically, as a resource to inform drug development from an organizational perspective and identify features that could be predictive of trial success.

The next in line.

This Review discusses the scientific and technological breakthroughs reshaping transplant medicine, with a focus on the transformative role of artificial intelligence, xenotransplantation and tissue engineering as new sources of organs, and novel approaches to immunosuppression.

The pandemic is being used as a pretext to push unproven artificial-intelligence tools into workplaces and schools.

A review of the literature on artificial intelligence systems to examine openness reveals that open AI systems are actually closed, as they are highly dependent on the resources of a few large corporate actors.

Chlorine electrosynthesis from seawater is limited by poor selectivity and stability under industrial-scale conditions. Here atomic-step-enriched ultrafine high-entropy alloy nanowires enable highly efficient chlorine evolution at 10 kA m⁻² for over 5,500 h through dynamic Pt–O active sites, reducing electricity consumption and feedstock costs for next-generation chlor-alkali processes.

An exemplary book gets much right, but misses the point that the nation’s tech dominance has more to do with globalization than with policy choices.

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.

Trained on large and multicenter datasets with different imaging modalities, a foundation model is shown to have strong performance on the full spectrum of clinically relevant tasks and to increase user accuracy in diagnostic tasks.

A new study combines MRI data with proteomic, metabolomic and RNA data to develop and examine seven organ-specific MRI-based aging clocks, uncovering links to overall mortality and organ-specific diseases.

The Mouse Cancer Cell line Atlas (MCCA) provides major advances towards a mechanistic understanding of cancer genomes.

Follow-up telephone interviews are crucial but labour intensive in large-scale clinical trials. Here, the authors show that a specialized large language model automates this process, outperforming both existing AI systems and human experts in accuracy and efficiency.

A specialized, open-source, retrieval-augmented language model is introduced for answering scientific queries and synthesizing literature, the responses of which are shown to be preferred by human evaluations over expert-written answers.