Embodied intelligence has enabled machines to act autonomously, yet intelligence remains largely confined to digital models, leaving robotic bodies as passive executors. In extreme environments, such ...

Reproducibility has long been a concern in machine learning research. The rise of large language models adds new layers of complexity, underscoring the need for clearer reporting standards and communi...

While there are widespread concerns about the potential negative societal impacts of algorithm-driven systems, a more balanced approach to understanding online content consumption and engagement that ...

Large language models can potentially serve as human proxies across various roles, enabling nuanced simulations of behavior. This Review categorizes their applications, stressing distinct evaluation m...

A descriptor, TDiMS, is developed to summarize the enumerated pairwise topological distances between molecular substructures and identify interpretable molecular features for materials discovery.

A natural-language-guided pathology segmentation model is developed to link pathological language with pathology image content to produce semantic masks, offering a path toward computational pathology...

This Resource introduces PathSegmentor—a foundation model that segments 160 pathological categories via natural language prompts, generalizes across datasets and clinical cohorts, and enables scalable...

MutexaGPT turns plain-language enzyme-engineering intuitions into lead mutation designs through AI agent-orchestrated high-throughput molecular modeling, yielding experimentally validated improvements...

The authors develop a physics-transfer learning framework that learns cortical folding physics from simple geometries and transfers it to complex brain structures, enabling accurate prediction of brai...

Nature Computational Science - AI testing is not AI safety

A large-scale experiment at a major machine learning conference shows that when researchers privately rank their own submissions, those rankings forecast future citations more reliably than peer-revie...

This work presents an efficient and modular framework called CrysVCD, which integrates chemical rules directly into the generative process for valence-constrained design.

Most patients leave the clinic with only a few of the clinical measurements needed to tell the full story of their health. PULSE is a new artificial intelligence (AI) framework that fills in the blank...

PULSE is an AI framework that uses patients’ longitudinal electronic health records to accurately generate within- and cross-modal data from routine clinical laboratory tests, enabling cost-effective precision medicine across diverse biomedical settings.

The study finds that authors’ rankings of their own AI conference papers predict later citations better than peer-review scores, suggesting that self-rankings could provide a simple complement to peer...

As AI becomes increasingly embedded in research and scientific publishing, transparency, accountability, and human oversight remain essential to safeguarding research integrity.

As large language models begin to plan and run experiments on their own, the reflex is to demand that they be interpretable before they are trusted. We argue that what grounds trust in autonomous scie...

Addressing generalization limits in machine learning models for chemical reactions, this work introduces a pretraining scheme that enables accurate predictions for unseen chemistry and expands computa...

A newly generated high-throughput experimentation (HTE) dataset is integrated with open-source data to produce a large and high-quality Buchwald–Hartwig reaction dataset for training machine learning ...

A predictive model capable of generalizing to previously unseen Buchwald–Hartwig chemical space is built to enable rapid, in silico optimization, reducing reaction scouting time from weeks to minutes.

SpatialFormer, a transformer-based model for spatial transcriptomics, learns cell–cell relationships to predict cell co-localization, annotate cell types and niches, and reveal cell communication gene...

A new method designs RNA sequences by learning from alignments of structurally similar molecules, producing aptamers and ribozymes that are experimentally validated.

This study presents AlignIF, which uses multiple structure alignments to capture evolutionary rules and conservation patterns across RNA families, enabling diverse, structure-guided designs of functio...

MIRACLE is an online continual learning framework that efficiently integrates single-cell multimodal data across batches, modalities, tissues, and diseases.

Privacy, utility, and efficiency benchmarks have advanced privacy-preserving genome-wide association studies (GWAS) but often fall short of supporting organizational decisions about responsible data s...

This study reports personalized models that decode human emotion states from intracranial brain recordings in real time, work across different tasks and uncover key brain networks that could guide fut...

We highlight how AlphaFold2 has advanced protein design and inspired the development of artificial intelligence-driven infrastructures for scientific discovery.

Universal machine learning force fields excel on computational benchmarks but consistently fail when tested against experimental measurements of real minerals, underscoring the need for evaluation fra...

A generative model designs information metamaterials from meta-atoms to programmable arrays, enabling beam steering, focusing and holography with experimental validation and more than 1,000-fold faste...

Protein language models are powerful computational tools for protein fitness prediction. The recent investigation into the relationship between model outputs and prediction accuracy reveals the scalin...

Larger language models do not always perform better at fitness prediction. Hou et al. found that performance depends on whether model outputs match evolutionary patterns in homologs, which is best ach...

With this Collection, the editors invite contributions that advance the foundations and applications of high-fidelity digital twins in drug discovery.

This study shows that local pseudopotentials boost the efficiency of neural network-based quantum Monte Carlo while preserving accuracy, enabling reliable simulations of large, challenging iron–sulfur...

A diffusion-based multimodal framework called DISSECT leverages transcriptomic gradients to refine segmentation boundaries, enabling high-precision single-cell segmentation even in densely packed regi...

The authors propose a framework that integrates cytological imaging and spatial transcriptomics to improve cell segmentation and enable reconstruction of spatial single-cell transcriptomes.

This study introduces DIPPER, a GPU-accelerated tool that efficiently reconstructs evolutionary trees from tens of millions of genome sequences, extending scalability by one to two orders of magnitude...

The researchers developed NEP89, a foundation model for large-scale molecular dynamics across 89 elements with near-quantum-level accuracy and high efficiency. It can be rapidly adapted to address com...

Organoid intelligence represents a new frontier for computing inspired by the human brain. This Review discusses the evolution of brain-inspired computing along with potential challenges for future de...

HUMID is a bottom-up coarse-grained framework for proton transport that bridges the quantum and molecular scales, enabling simulations of reactive transport dynamics across diverse environments with o...

This study introduces RF-PHATE, a supervised visualization method that preserves data structure while revealing biological patterns, enabling insights into disease progression in multiple sclerosis, C...

Although social technologies are increasingly co-shaping the public sphere, these systems were not designed as democratic infrastructure. Here, we propose a framework for societal alignment that focus...

Geographical disparities regarding the availability of GPU hardware are becoming a structural constraint on scientific participation itself and must be addressed.

A new library of combinatorial optimization problems provides a standardized testing ground to rigorously compare quantum and classical algorithms, paving the way for demonstrating quantum utility.

This Resource presents the Quantum Optimization Benchmarking Library, which enables fair, reproducible benchmarks of quantum heuristics for ten difficult combinatorial optimization classes with baseli...

This study develops enhanced quantum solvers for an NP-complete Boolean satisfiability problem and reports empirical scaling advantages over leading classical solvers in large-scale simulations, with ...