Machine learning articles from across Nature Portfolio

We trained a three-dimensional visual foundation model directly on 5.24 million routine clinical computed tomography (CT) and magnetic resonance imaging (MRI) image series so that it learned a shared representation of neuroanatomy and disease. The model demonstrated state-of-the-art diagnosis, in contrast to foundation models that are trained on public Internet and medical data, and enabled preliminary report generation and triage in real health systems.

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

The development of two medical AI assistants highlights an unnerving challenge: as the technology races ahead, what is the best way to evaluate what works?

Predicting future metastases remains a major clinical challenge. Here, the authors develop EmitGCL, a deep-learning framework to predict metastasis and related biomarkers using cancer single-cell sequencing data, enabling and validating the discovery of occult metastases and breast cancer metastasis biomarkers.

Integrating single-cell RNA-sequencing (scRNA-seq) data across species is still technically challenging. Here, the authors report a transfer learning framework designed to integrate scRNA-seq data across species by combining RNA expression with embeddings from protein language models and general-purpose language models.

Atrial cardiopathy precedes atrial fibrillation and stroke, yet it is detected through costly, inaccessible cardiac imaging. Here, the authors show that a deep learning model estimates left atrial structure and function from a 12-lead ECG and independently predicts adverse cardiovascular outcomes.

Researchers develop a machine learning model that predicts how protein phase separation responds to environmental conditions. It is expected to help pinpoint pathological phase separation and design protein condensates with desired behavior.

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

We trained a three-dimensional visual foundation model directly on 5.24 million routine clinical computed tomography (CT) and magnetic resonance imaging (MRI) image series so that it learned a shared representation of neuroanatomy and disease. The model demonstrated state-of-the-art diagnosis, in contrast to foundation models that are trained on public Internet and medical data, and enabled preliminary report generation and triage in real health systems.

Research suggests that the combination of incentives to publish and the use of large language models will lead to more papers, but they will be less refined.

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

The model can modify existing proteins using a probability approach.