Wang et al. build a chromatin accessibility atlas of 650 gastric tissue samples spanning pre-cancer to advanced cancer, revealing a cancer-like epigenetic priming state in metaplasia that licenses tumor initiation. Advanced cancers segregate into three epigenomic subtypes; the pseudo-gastric subtype resists immunotherapy via FOSL2-driven H+/K+ ATPase expression that acidifies the microenvironment and suppresses CD8+ T cells—a phenotype reversed by the acid blocker vonoprazan.

In a phase 1 trial, inhibition of the Werner syndrome helicase (WRN) enzyme exploits a new synthetic lethal vulnerability in cancers with microsatellite instability, offering a new way to target these tumors beyond immune checkpoint inhibition.

Nature - An AI revolution is sweeping through education. But teachers who can convey the excitement of thinking and learning are irreplaceable.

Nature - There is growing concern that AI can blunt memory and reasoning. But science shows ways to keep the brain sharp.

Deep learning for single-cell Gene Regulatory Networks - aertslab/deepSCENIC

Nature - Using a video game involving bombs and treasure chests, researchers show how neural activity flickers between two regions of the frontal cortex when the brain weighs up a tricky decision.

Scaffolded DNA Computer performs diverse computations while thermodynamically relaxing towards equilibrium, demonstrating simplicity, reusability, speed, robustness and scalability of the approach by running 10 programs, including parity, multiplication and 25-bit addition.

Artificial intelligence is poised to reshape clinical trials, addressing challenges in design, recruitment, monitoring and conduct. This Review outlines a framework built on deeply phenotyped cohorts, validated surrogate end points, digital twins and agentic AI systems that could shorten timelines and support the trial life cycle, highlighting opportunities to accelerate drug development while maintaining rigorous regulatory and benefit–risk assessment.

A study shows that integrating numerous multimodal, patient-level biomarkers enhances the prediction of cancer immunotherapy response, but generalizability is still a challenge.

Nature Reviews Molecular Cell Biology - Single-molecule genomics methods are used to study the activity of regulatory factors on individual DNA molecules genome-wide, thereby enabling...

The genomic regions regulating gene expression are often themselves transcribed into a variety of noncoding RNAs (ncRNAs). However, the regulatory roles of this noncoding transcription remain largely ...

The transition from AI-assisted tools to more integrated, AI-native models of care requires workflow integration, clinician engagement and measurable clinical value.

Luo et al. develop ELF, a framework that ensembles five pathology foundation models to generate unified representations from whole-slide images. Trained on nearly 54,000 slides across 20 anatomical sites, ELF outperforms individual foundation models in disease classification, biomarker detection, and prediction of response to chemotherapy, targeted therapy, and immunotherapy.

Scientists are probing whether language models trained on historical data can reproduce creative breakthroughs.

Chen et al. present UniCure, a multi-modal foundation model that predicts drug-induced transcriptomic responses across diverse cellular contexts. By bridging cancer cell line data with patient-derived models, UniCure enables personalized drug prioritization and patient stratification, with its predicted therapeutic rankings correlating strongly with clinical survival outcomes in real-world cohorts.

AI model AlphaGenome forecasts the consequences of altering every single DNA letter in the human genome.

An expanded atlas of human cells maps the expression of nearly all human transcription factors across 701 cell-type/tissue combinations and uncovers the heterogeneous nature of senescence across tissues and cell types.

Concept-Wrapper Network, a concept-based explanation approach deployed on a real autonomous vehicle, improves drivers’ ability to understand and predict its behaviour.

Malva enables ultrafast, reference-free searching of raw single-cell sequences across millions of cells, transforming static transcriptomic atlases into dynamic resources for understanding the functions of sequences.

Metastatic dissemination represents the major determinant of poor clinical outcome across cancer entities. Yet, how driver oncogenes shape transcriptional programs facilitating metastasis is poorly un...

PULSE is an AI framework that uses patients’ longitudinal electronic health records to accurately generate within- and cross-modal data from routine clinical laboratory tests, enabling cost-effective precision medicine across diverse biomedical settings.

High-dimensional data are visualized through representation on tree structures.

The authors outline strategies to maximize the impact of ADCs by integrating multiple chemistry advances into drug constructs, and developing infrastructures and tools to accelerate and de-risk the preclinical and early clinical development of these agents.

Nature Reviews Cancer, Published online: 25 August 2026; doi:10.1038/s41568-026-00969-1In this Review, Franceschini and colleagues discuss how remodelling of the 3D structure of chromatin in cancer is both a cause and a consequence of genetic and epigenetic alterations and highlight how chromatin 3D conformation might be exploited for diagnosis and therapy.

Company Description: UHN is Canada’s #1 hospital and the world’s #1 publicly funded hospital. With 10 sites and more than 44,000 TeamUHN members, UHN consists of Toronto General Hospital, Toronto West...

Dr. Dan-Avi Landau, a world-renowned cancer researcher and pioneering geneticist, has been appointed inaugural chair of the newly created Department of Systems and Computational Biomedicine. The new ...

A pilot implementation study of a large language model clinical decision support system in a tertiary emergency department showed safe integration but steadily declining clinical adoption over 4 weeks.

Hyeon et al. integrate protein structures with population-scale cancer genomics data to characterize significantly mutated regions in 173 cancer-relevant genes. These regions align with functional domains, three-dimensional structural constraints, and gene-specific biology, providing a framework for classifying mutation patterns in cancer genes and prioritizing sparsely annotated variants for clinical interpretation.

The authors outline strategies to maximize the impact of ADCs by integrating multiple chemistry advances into drug constructs, and developing infrastructures and tools to accelerate and de-risk the preclinical and early clinical development of these agents.

Drawing inspiration from Hilbert’s list of 23 mathematical problems that have focused the mathematical community’s attention for more than a century, we propose fifteen grand AI challenges to focus the biomedical community’s attention on critically relevant questions, most of which still lack effective predictive methodologies.

Manipulating biology in the physical world is complex; the authors envision an AI-driven digital organism—a system of integrated multiscale foundation models—detailing how to construct and use such a model to advance biological and medical research.