Identifying a sparse, reliable set of biomarkers from large multi-omics datasets: Hédou et al. present a machine learning framework that selects the most reliable biomarkers from large multi-omics datasets. By injecting artificial noise (gray dots) into a dataset containing thousands of omics features (pink dots), Stabl calculates reliable predictive biomarkers (circled dots).

See Hédou et al.

A representation of cellular mRNA modified by m6A, which is depicted as a gemstone. Marayati et al. develop GEMS, a genetically encoded m6A sensor that can report changes in m6A in living cells.

See Marayati et al.

An illustration of liposomal antibiotics (pink lipid particles) targeting tumor-associated bacteria (colored spheres in central purple tumor), leading to activation of T cells (white) by cytokines (orange) released from antigen-presenting cells (not shown). Wang et al. show that killing bacteria in tumors by liposomal delivery of antibiotics generates neoantigens that enhance T cell antitumor immunity in a mouse model of colon cancer.

See Wang et al.

A DNA nanodevice measuring organellar sodium ion levels in lysosomes of living cells. Zou et al. developed RatiNa for imaging intracellular Na+ at single-organelle resolution.

See Zou et al.

Predicting the outcome of genetic perturbations: Roohani et al. present GEARS, a computational method that integrates deep learning with a knowledge graph of gene–gene relationships to simulate the effects of genetic perturbations.

See Roohani et al.

An illustration of transplanted human glial progenitor cells (red) killing resident human glia (pink) in the adult mouse brain. Vieira et al. show that transplanted young glial progenitors outcompete older or diseased glia, suggesting a therapeutic strategy for neurodegenerative and myelin diseases.

See Vieira et al.

Image of a whole mouse body with depth color-coding of a neuronal marker. Mai et al. present wildDISCO, a method for whole-body imaging compatible with standard IgG antibody labeling.

See Mai et al.

A miniaturized ultrasonic system can provide full-body monitoring and diagnosis. Lin et al. engineer a soft, wireless ultrasound device to measure deep body signals in moving subjects, offering a hands-free solution for continuous, comprehensive health tracking at the point of care.

See Lin et al.

Artistic impression of the three data types key to machine learning for functional protein design: structure, sequence and labels. The structure of carbonic anhydrase is shown in front of a background composed of amino acid letters representing sequence data. The hue overlay represents a fitness landscape that experimentally acquired labels help to map.

See Notin et al.

Neurons in the mouse hippocampus imaged with the Schmidt objective. Voigt et al. develop a multi-immersion objective based on a mirror instead of a lens.

See Voigt et al.