Reviews & Analysis

We developed 4Pi-SIM, which integrates 4Pi microscopy with structured illumination microscopy to achieve isotropic optical resolution through interference in both illumination and detection. We demonstrate its capabilities by time-lapse volumetric imaging of various subcellular structures at 100-nm resolution.

We provide scalable new tools, which include mosaic models and downstream multiplexing, for longitudinal single-cell profiling of brain organoids derived from multiple individuals. In addition, we benchmark computational methods for cell identity deconvolution and provide guidelines on how to use each multiplexing method for charting neurodevelopmental trajectories.

This Review provides a comprehensive and detailed discussion about how to build and use single-cell atlases.

Graphene energy transfer (GET) can be used to measure the distance between a graphene surface and a fluorescent dye. A new strategy for attaching DNA in a controlled orientation onto graphene now makes it possible to use GET to track the dynamics of DNA and protein–DNA interactions with base-pair resolution.

This Review synthesizes the literature from over 10 years of Cell Painting for image-based profiling and highlights how advances in this technology enable new biological discovery of cellular phenotypes and cell responses to perturbations.

We developed scAtlasVAE, a deep-learning-based method to integrate large-scale single-cell RNA sequencing data. Using this tool, we established a comprehensive, million-scale, pan-disease human CD8⁺ T cell reference atlas, which incorporates information on transcriptomes and the diversity of the αβ T cell receptor repertoire.

Large language models (LLMs) demonstrate potential as assistants in functional genomics, offering a new avenue for gene set analysis. In our evaluation of five LLMs, GPT-4 was the top-performing model and generated common functions for gene sets with high specificity, reliable self-assessed confidence and supporting analysis, complementing traditional functional enrichment.

The Nucleotide Transformer is a series of foundation models pre-trained on DNA sequences through self-supervised learning that extracts context-specific representations of nucleotide sequences. These representations can then be used to accurately predict molecular phenotypes.

We present an RNA language model-based deep learning pipeline for accurate and rapid de novo RNA 3D structure prediction, demonstrating strong accuracy in modeling single-stranded RNAs and excellent generalization across RNA families and types while also being capable of capturing local features such as interhelical angles and secondary structures.

Massively parallel base and prime editing screens provide unparalleled interrogation of genetic variants with single-nucleotide resolution.

A new deep learning approach enables motion correction across sequential acquisitions, even if each image in the sequence was acquired differently.

A groundbreaking biomedical AI foundation model, called BiomedParse, unifies detection, segmentation and recognition of organs, setting the stage for enhanced efficiency and accuracy in biomedical research and diagnostics.

This Perspective highlights the need to develop methods for single-molecule temporal omics studies and discusses nanopores as a potential solution, as well as the challenges associated with using nanopores for the analysis of complex biological samples.

Two independent studies provide comprehensive human embryo reference maps by integrating multiple human embryo single-cell RNA sequencing (scRNA-seq) datasets. These references are instrumental in advancing cell type annotation and benchmarking stem cells and stem cell–based embryo models.

This Perspective analyzes the most common maturation and assessment techniques for in vitro hiPSC-derived cardiomyocyte models and makes recommendations for standardizations in this field.

A Bayesian inference framework improves RNA velocity estimates to make them more coherent and biologically interpretable.

The accuracy of SCUBA-D, a protein backbone structure diffusion model trained independently and orthogonally to existing protein structure prediction networks, is confirmed by the X-ray structures of 16 designed proteins and a protein complex, and by experimental validation of designed heme-binding proteins and Ras-binding proteins.

Cryo-electron microscopy with energy resolution, using the EELS-STEM method, allows researchers to identify the approximate locations of certain heavier atoms within single frozen, hydrated protein particles.

Pangenome graphs signify a new frontier in genome representation. Recent advances in constructing and personalizing them mark progress in this area.

We developed LABEL-seq, a platform that enables measurement of protein properties and functions at scale by leveraging the intracellular self-assembly of an RNA-binding domain (RBD) and protein-encoding RNA barcode. Enrichment of RBD–protein fusions, followed by high-throughput sequencing of the co-enriched barcodes, enables the profiling of protein abundance, activity, interactions and druggability at scale.