Volume 9 Issue 3, March 2025

Generating low-dimensional latent spaces for gene-expression data via unsupervised deep learning can unveil biological insight across cancers.

A modular model integrating clinical metadata and mammography and tri-modal ultrasound images from patients presenting to the clinic with breast cancer symptoms performs similarly or better than experienced human experts at differential diagnosis and tumour classification.

The inference process of medical-image classifiers can be audited by levering the expertise of physicians to identify medically meaningful features in ‘counterfactual’ images produced via generative AI.

A deep-learning model can accurately quantify circulating tumour DNA from the density distribution of cell-free DNA-fragment lengths in plasma from patients with cancer and from healthy individuals.

Cascaded diffusion models can be used to synthesize realistic whole-slide image tiles from latent representations of RNA-sequencing data from human tumours.

Using unsupervised deep learning to generate low-dimensional latent spaces for gene-expression data can unveil biological insight across cancers.

A multimodal model for the stratification of breast cancer risk based on clinical metadata, mammography and trimodal ultrasound images performed as well as or better than radiologists at tumour classification and at differential diagnosis.

An interpretable transformer-based model leveraging graph representation learning accurately predicts cancer genes across homogeneous and heterogeneous pan-cancer networks of biological interactions.

Minimal tumour perturbations that boost T-cell infiltration can be discovered by using deep learning to analyse large-scale spatial omics profiles of tumours.

Self-supervised representation learning on data from imaging mass cytometry can be used to distinguish morphological differences in tumour microenvironments and to precisely characterize distinct microenvironment signatures.