Search

Foundation models enable rapid adaptation to various downstream tasks and are hence about to become a new paradigm in biomedicine. Here, the authors develop LLaVA-Rad, a small AI that bests larger models in chest X-ray interpretation, and CheXprompt, a radiologist-aligned factuality metric, to enable scalable, privacy-preserving image analysis.

Generation of mature oocytes requires tight regulation of a discontinuous meiotic cell cycle. Here they show that the acetyltransferase Nat10 mediates modification of RNAs targeted for degradation and find that this process is essential for female oocyte meiosis and maturation.

The Airy-Talbot effect is experimentally demonstrated for spoof surface acoustic waves in a structured metasurface. Owing to its self-imaging and self-healing properties, the authors achieve robust multipath transmission of nonperiodic signals.

The distinction of liver lesions is critical for the accurate diagnosis and treatment of liver cancer. Here, the authors develop LiLNet, a deep learning-based system to identify focal liver lesions as well as benign and malignant liver tumours from CT images with high accuracy across multiple patient cohorts.

Diffusion models have shown promise in content generation, however, like many forms of generative AI, this comes with high computation cost, exacerbated by standard von Neuman computing architectures. Here, Cheng et al present a magnetoelectric memory for in-memory computing and demonstrate diffusion-based image generation on an 80×80 array.

The pathological identification of lymph node metastasis in whole-slide images is demanding and tedious. Here, the authors design an artificial-intelligence-assisted assessment workflow to facilitate the routine counting of metastatic LNs.

Long-circulating, transfection-competent LNP-mRNA systems are key for effective extrahepatic delivery. Here, authors show that LNPs with high bilayer lipid ratios yield high mRNA encapsulation, prolonged circulation, and enhanced transfection in extrahepatic tissues.

An integrated dataset combining genetics, epigenomics, transcriptomics, proteomics and metabolomics from 1,342 people living with HIV illuminates molecular pathways driving immune responses and comorbidities in this population.

Yandao Huang and colleagues introduce a non-contact system that integrates fiber optic sensors with AI to achieve accurate, medical-grade ballistocardiography signal detection. This system allows for continuous nocturnal blood pressure monitoring, aiding in early screening and managing hypertension and other cardiovascular diseases.

Copper-based organic light-emitting diodes are low-cost alternatives to precious metal-based counterparts. Here, the authors develop copper(I) thermally activated delayed fluorescent emitters with colors spanning from blue to near infrared, and achieve device lifetime close to 3700 h at 1000 cd/m².

OmiCLIP is a visual–omics foundation model that integrates histology and spatial transcriptomics. The associated Loki platform offers accurate and robust tools for alignment, annotation, cell-type decomposition and spatial gene expression prediction.

When it comes to reasoning about the motion of physical objects, humans have natural intuitive physics knowledge. To test how good artificial learning agents are in similar predictive abilities, Xue and colleagues present a benchmark based on a two-dimensional physics environment in which 15 physical reasoning skills are measured.

Spatiotemporal beams differ from standard spatial structured light. The authors introduce 2D spacetime duality to uncover the conventional–spatiotemporal beam connection, enabling the development of high-fidelity spatiotemporal beams.

It is still challenging to make accurate diagnosis of biliary atresia (BA) with sonographic gallbladder images particularly in rural areas without relevant expertise. Here, the authors develop a diagnostic deep learning model which favourable performance in comparison with human experts in multi-center external validation.

Inherited retinal diseases are both numerous and diverse, but all arise from genetic mutations leading to retinal degeneration. Through the use of modern diagnostic tools, accurate genotype prediction is now possible using high-resolution imaging techniques alone, facilitating improved screening and genetic variant prioritization.

Efficient and accurate constant potential simulation is essential for electrochemical modeling. Here, authors propose a constant potential framework by combining the machine learning force field with the charge equilibration method to study the dendrites nucleation at the Li metal battery interface.

Here, the authors introduce Photoacoustic Tomography with Temporal Encoding Reconstruction (PATTERN) - a high-speed, non-destructive photoacoustic brain imaging technique that constructs 3D fluorescent maps of the brain and improves upon some of the limitations associated with traditional whole-brain optical imaging techniques.

The fungal cell wall is a complex structure composed mainly of glucans, chitin and glycoproteins. Here, the authors use solid-state NMR spectroscopy to assess the cell wall architecture of Aspergillus fumigatus, comparing wild-type cells and mutants lacking major structural polysaccharides, with insights into the distinct functions of these components.

Polygenic scores for body mass index and obesity constructed using data from 5 million people with different ancestries were associated with distinct weight gain trajectories from early childhood to adulthood.

FastGlioma is a visual foundation model for fast and accurate detection of glioma infiltration in fresh, unprocessed surgical tissue.

Wang et al. present E3-CryoFold, a deep learning method for cryo-EM structure determination that enables end-to-end training and one-shot inference This method reduces inference times while boosting template modelling scores against comparable methods.

Drug-induced interstitial lung disease (DIILD) is emerging as a hidden toll of modern oncology. Diagnosis of DIILD is confounded by non-specific symptoms and the absence of validated biomarkers, leaving clinicians to navigate heterogeneous, drug-specific guidelines. Herein, we discuss this predicament and advocate a shift from reactive management towards proactive survivorship.

Electrocardiogram is crucial for cardiovascular disease prevention, diagnosis, and prognosis. Here, the authors present a motion-unrestricted dynamic electrocardiogram system utilizing skin-conformal, imperceptible electronics for long-term, comfortable, and accurate 12-lead monitoring.

Tailored to provide diabetes management recommendations from large training and validation datasets, an artificial intelligence system integrating language and computer vision capabilities is shown to improve self-management of patients in a prospective implementation study.

Based on a newly developed deep-learning model for predicting the binding affinities of protein–protein interactions, the effects of SARS-CoV-2 spike protein variants on binding to the ACE2 receptor and neutralizing antibodies were analyzed and used to predict immune escape and viral evolution.

Xue et al. develop a foundation model for spectral analysis. It excels in multiple tasks, such as denoising spectra and quantifying trace molecules, and is especially promising in identifying metabolic biomarkers of diseases such as stroke, Alzheimer’s disease and prostate cancer.

Machine learning can be used to infer proper retrosynthesis routes for newly designed molecules. Here, the authors develop a multitask graph representation learning model for single-step retrosynthesis inference by exploiting chemical synthesis rules among different entity.

Uptake of explainable artificial intelligence (XAI) methods in geoscience is currently limited. We argue that such methods that reveal the decision processes of AI models can foster trust in their results and facilitate the broader adoption of AI.

Very few of the COVID-19 ML models were fit for deployment in real-world settings. In this Comment, Huang et al. discuss the main steps required to develop clinically useful models in the context of an emerging infectious disease.

ROAM, based on large regions of interest and a pyramid transformer, can automatically capture key morphological features consistent with the experience of pathologists to provide accurate, reliable and adaptable clinical-grade diagnoses of gliomas while advancing the discovery of molecular and morphological markers related to glioma diagnosis.

MGAT1 is a glycosyltransferase critical for the synthesis and maturation of complex N-glycans. Here, the authors show that MGAT1 modulation of CD73 glycosylation and function regulates tumor immune response in triple-negative breast cancer.

Photoreceptors are specialized neurons of the retina that are sensitive to systemic metabolic stresses. Here, the authors combine in vivo photoreceptor assessment with multisystem health readouts and circulating metabolomics to predict multi-disease risk.

Manual postprocessing of computed tomography angiography (CTA) images is extremely labor intensive and error prone. Here, the authors propose an artificial intelligence reconstruction system that can automatically achieve CTA reconstruction in healthcare services.

Zeolite-confined metals are valuable catalysts, but finding a stable metal-zeolite pair with a simple synthesis is challenging. Here, the authors use data-driven simulations to guide the one-pot synthesis of a stable Pt1@Ge-MFI catalyst, achieving high activity, selectivity, and durability in propane dehydrogenation.