Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain
the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in
Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles
and JavaScript.
After years of progress, density functional theory is entering a period of rapid advancement, enabled by emerging generalized schemes, richer descriptors, machine learning, and the anticipated development of broader, higher-quality datasets.
Submitting an appeal regarding an editorial decision may require a significant investment of time and effort from authors. Therefore, it is important to understand what an appeal entails before making the decision on whether to appeal.
The authors demonstrate that the accuracy of predictions in network biology scales with larger foundation models pretrained with larger, more diverse data and that quantization enables resource-efficient predictions while preserving biological knowledge.
HorusEye is a foundation model for universal X-ray tomography restoration that learns realistic degradation directly from data. It supports imaging at substantially lower doses and reduces hardware requirements while improving expert analysis and downstream AI performance.
InstructNA leverages nucleic acid large language models with HT-SELEX for de novo generation of functional nucleic acids, exhibiting high efficiency and general applicability in designing aptamers for various targets.
This study introduces a unified framework for brain MRI tissue segmentation and region parcellation across the lifespan, demonstrating robust and consistent performance across heterogeneous datasets using a single model.
A reward function (TANGO) is developed to enforce building blocks in generative artificial intelligence and leverage the synthesizability of high-value materials.
PoTS is an automated pipeline that maps reaction transition states inside zeolite pores. By identifying hundreds of confined transition states across many frameworks, it explains differences in catalytic selectivity and informs zeolite design.
A machine learning framework reveals how dynamic routing and interpretability can accelerate the discovery of better electrolytes for next-generation batteries.
The PropMolFlow model uses flow matching to efficiently generate chemically valid molecules in three dimensions with targeted properties, enabling accelerated discovery of molecules useful in materials and pharmaceutical science.
The authors propose the TANGO reward function, which enables the generation of property-optimized small molecules with predicted synthesis routes, incorporating a small set of shared precursors.
We developed a dual-module neural network, CATS Net, that models how the human brain compresses sensorimotor experiences into abstract concepts. CATS Net activation patterns aligned with those associated with concept formation and understanding in the human brain. The model also enabled conceptual communication between artificial agents without human language.
Large-scale cooperation is characterized by complex interaction patterns with nonlinear outcomes. Deepening our understanding may be critical to addressing real-world collective challenges.
APOLLO is an autoencoder-based framework to integrate diverse data modalities while preserving both shared and modality-specific information. It enables predicting missing data modalities and identifying the influence of each modality on a phenotype.
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.
This work introduces an unsupervised method that restores high-quality Raman hyperspectral images from low-light measurements, enabling faster, lower-power imaging and expanding the use of Raman techniques in biological and chemical analysis.
This study presents a dynamic routing-guided framework to model and interpret salt–solvent chemistry, which effectively handles long-tailed data and captures the full spectrum of formulations, shaping the conductivity atlas for non-aqueous electrolytes.
The CATS Net framework models how abstract concepts emerge from sensory experience. Aligning with human brain activity and enabling knowledge transfer, it provides a unified framework for understanding conceptual intelligence in both humans and AI.
By learning directly from atomic motion, without the need for handcrafted descriptors, a graph neural network reveals how molecular systems change state, delivering accurate kinetics and atom-level insight into complex transitions.
