Reviews & Analysis

A new sequencing-based method optimizes barcode-based synaptic mapping by capturing paired neuronal barcodes from intact synaptic fragments, achieving the efficiency needed to scale toward comprehensive brain connectivity maps.

Language model-inferred embeddings are replacing structure-derived descriptions of proteins, genes and genomes. We propose a model-agnostic measure to quantify reliability of these new representations.

Sensing and mapping lipid species in living cells has been challenging. Here, we show that hyperspectral fingerprint opto-acoustic microscopy (HyFOPM) achieves label-free visualization of specific lipid species, including sphingomyelin and cholesterol, within living cells. This technology could offer insight into lipid metabolism and disease without requiring fluorescence or other exogenous tags.

Tunable hydrogel-based micropillar arrays are fabricated to model the biomechanical cues of size, shape and rigidity characteristic of axons. These next-gen micropillars allow the exquisite examination of the cell biology of oligodendroglia, resolving the myelinogenic potential, myelin extent and kinetics of wrapping in a cell-autonomous manner.

By allowing protein language models (PLMs) to learn from each other’s most confident predictions, we compressed the collective knowledge of existing PLMs into VESM — a single sequence-only model that outperforms state-of-the-art hybrid methods. VESM predictions extended beyond binary pathogenicity classification, accurately quantifying the severity of variant effects on clinical phenotypes.

This Perspective provides recommendations for open and sustainable AI and associated resources facilitating exploration and implementation.

A new AI framework autonomously explores single-cell RNA sequencing datasets and produces creative, biologically relevant findings.

Making live mammalian tissues transparent for imaging experiments without compromising their normal cellular functions has been a long-standing challenge. An isotonic and minimally invasive optical clearing medium for live mammalian cells and tissues (named SeeDB-Live) paves the way for deep-tissue live imaging of cellular functions ex vivo and in vivo.

Annevo uses deep learning to achieve unprecedented accuracy in eukaryotic gene annotation, approaching the performance of evidence-based methods.

We have engineered symmetric RNA scaffolds that allow structure solution of small RNAs via single-particle cryogenic-sample electron microscopy (cryoEM). This method enabled small-molecule ligand placement in experimental maps and revealed the molecular basis for specificity of natural and synthetic RNA aptamers, thus suggesting routes for structure-guided RNA engineering.

This Perspective establishes a comprehensive and practical framework to guide intrinsically disordered protein (IDP) ensemble determination, benchmarking and interpretation, as well as proposes a roadmap for IDP ensemble determination, uncertainty quantification and actionable benchmarking strategies.

This study introduces an unprecedented resource of thousands of 3D images representing the diversity of an ecologically dominant group of insects: ants. These data were generated using high-throughput synchrotron X-ray microtomography, which opens a pathway to scalable digitization of the vast diversity of invertebrate life.

The new algorithm DECODE transcends the limitations of omics-specific methods, and provides a unified framework for the deconvolution of transcriptomic, proteomic and metabolomic datasets.

Current single-cell metabolomics methods show low sensitivity and limited coverage of small-molecule metabolites. We developed an ion mobility-resolved mass cytometry technology that incorporates selective ion accumulation and cell superposition strategies to deliver high sensitivity and deep coverage, which captured over 5,000 metabolic peaks and about 800 metabolites from individual cells in a high-throughput manner.

Complex digital representations of organs were reconstructed by computationally generating virtual slices from sparsely sampled planar spatial transcriptomic data, exemplified by a 38-million-cell mouse brain atlas that bridges gaps between tissue sections and preserves the continuous three-dimensional (3D) molecular landscape at single-cell resolution.

We introduce TransBrain, a computational framework for bidirectional translation of whole-brain phenotypes between humans and mice. TransBrain enables quantitative cross-species comparison in a unified latent space and facilitates functional modeling of the human brain in mouse models.

This Review provides a comprehensive discussion of how methods in machine learning and computer vision have been used to improve super-resolution microscopy to gain insights into subcellular biology.

We developed SmartEM, a method that integrates machine learning directly into the image acquisition process of an electron microscope. By allocating imaging time in a specific manner — scanning quickly at first, then rescanning only critical areas more slowly — we are able to accelerate the mapping of neural circuits up to sevenfold without sacrificing accuracy.

This Perspective envisions the future of 3DTFM and provides practical guidelines for experimental and computation strategies for its application.

DynamicAtlas is a new open-source tool for incorporating gene expression and tissue shape changes into a single atlas with a continuous developmental timeline.