Search

Training machine learning models to predict the function of proteins is limited by the availability of only a small amount of labelled training data. Training can be improved by employing generative adversarial networks to generate additional synthetic protein samples.

High quality hit identification remains a considerable challenge in de novo drug design. Here, the authors train a generative adversarial network with transcriptome profiles induced by a large set of compounds, enabling it to design molecules that are likely to induce desired expression profiles.

Maurizi et al. introduce GraphMetaMat, a graph-based AI framework for designing 3D metamaterials with programmable nonlinear responses, enabling the inverse design of new structural and acoustic behaviours despite fabrication defects and limits.

Biological research relies on observing cell phenotypes, often obscured by natural variability. Here, the authors use generative modelling to unveil hidden changes triggered by infections, mutations, or drugs, allowing for accessible discovery of biomarkers.

The use of deep learning in medical image segmentation is limited by the low availability of annotated images. Here, the authors develop GenSeg, a generative deep learning framework that can generate high-quality paired segmentation masks and medical images that can improve the performance of segmentation models under ultra low-data regimes across multiple scenarios.

The authors used deep metric learning to characterize 650 neuroactive compounds by zebrafish behavioral profiles. After redesigning a large screen to overcome AI/ML shortcut learning, zebrafish behavioral similarity found compounds acting on the same human receptors as structurally dissimilar drugs.

From the blood immunome of 1,001 individuals aged 8–96 years, the authors used deep learning to develop an inflammatory clock of aging (iAge) that tracks with multimorbidity, immunosenescence, frailty and cardiovascular aging, and is also associated with exceptional longevity in centenarians. The main contributor to iAge is the chemokine CXCL9, which is shown to control endothelial cell senescence and function.

Performing multiple histological stains on a biopsy can be costly and time consuming. Here the authors present a method for the digital transformation of H&E stained tissue into special stains (e.g., PAS, Masson’s Trichrome and Jones silver stain), and demonstrate that it improves diagnoses over the use of H&E only.

Alzheimer’s disease is heterogeneous in its neuroimaging and clinical phenotypes. Here the authors present a semi-supervised deep learning method, Smile-GAN, to show four neurodegenerative patterns and two progression pathways providing prognostic and clinical information.

Phytochrome photoreceptors are master regulators of plant development. This paper describes 3D structures of soybean phytochrome A in both Pr (inactive) and Pfr (signalling) states, revealing changes that might transmit the light signal to the cell.

Many diseases can display distinct brain imaging phenotypes across individuals, potentially reflecting disease subtypes. However, biological interpretability is limited if the derived subtypes are not associated with genetic drivers or susceptibility factors. Here, the authors describe a deep-learning method that links imaging phenotypes with genetic factors, thereby conferring genetic correlations to the disease subtypes.

Deep Visual Proteomics combines machine learning, automated image analysis and single-cell proteomics.

Deep learning models were used to design synthetic cell-type-specific enhancers that work in fruit fly brains and human cell lines, an approach that also provides insights into these gene regulatory elements.

scBERT, a pretrained neural network for single-cell sequencing tasks, was published last year in Nature Machine Intelligence. To test the reusability of the method, Khan et al. use the code to assess the generalizablility of transformer architectures on single-cell genomics tasks.

Increase in clinical digital data is propelling the development and application of artificial intelligence methods in histopathology. In this Review, machine learning algorithms and models and their clinical use cases are discussed, highlighting the computational and operational challenges in the field.

Design of de novo synthetic regulatory DNA is a promising avenue to control gene expression in biotechnology and medicine. Here the authors present EspressionGAN, a generative adversarial network that uses genomic and transcriptomic data to generate regulatory sequences.

Deep learning can be used to virtually stain autofluorescence images of unlabelled tissue sections, generating images that are equivalent to the histologically stained versions.

Although the complexity of quantum systems scales exponentially with their size, classical algorithms and optimization strategies can still play an important role in the characterization of quantum states, their dynamics and the detection process.

Owing to the growing volumes of data from high-energy physics experiments, modern deep learning methods are playing an increasingly important role in all aspects of data taking and analysis. This Review provides an overview of key developments, with a focus on the search for physics beyond the standard model.

JWST observations of CH₃⁺ in a protoplanetary disk in the Orion star-forming region are reported showing that gas-phase organic chemistry in the interstellar medium is activated by ultraviolet irradiation and the methyl cation.

Deep space exploration missions will require new technologies that can support astronaut health systems, as well as biological monitoring and research systems that can function independently from Earth-based mission control centres. A NASA workshop explored how artificial intelligence advances could help address these challenges and, in this second of two Review articles based on the findings from the workshop, the intersection between artificial intelligence and space biology is discussed.

This Review discusses recent progress in bioinspired nanocomposite design, emphasizing the role of hierarchical structuring at distinct length scales to create multifunctional, lightweight and robust structural materials for diverse technological applications.

Weakly polar XH/π interactions are thought to be capable of influencing both the structure and function of proteins, but such interactions are usually identified from three-dimensional structural models. Now, using NMR spectroscopy and isotopic labelling, it has been shown that individual methyl/π interactions can be detected directly in proteins by measuring weak scalar couplings between the nuclei involved.

A deep learning framework, termed Fourier Imager Network (FIN), performs end-to-end phase recovery and image reconstruction from raw holograms, achieving unprecedented success in generalization to new types of samples.