Despite the importance of imaging in biological and medical research, a large body of informative and precious image data never sees the light of day. To ensure scientific rigor as well as the reuse of data for scientific discovery, image data need to be made FAIR (findable, accessible, interoperable and reusable). Image data experts are working together globally to agree on common data formats, metadata, ontologies and supporting tools toward image data FAIRification. With this Perspective, we call on public funders to join these efforts to support their national scientists. What researchers most urgently need are openly accessible resources for image data storage that are operated under long-term commitments by their funders. Although existing resources in Australia, Japan and Europe are already collaborating to enable global image data sharing, these efforts will fall short unless more countries invest in operating and federating their own open data resources. This will allow us to harvest the enormous potential of existing image data, preventing substantial loss of unrealized value from past investments in imaging acquisition infrastructure.

Identifying cell types in highly multiplexed images is essential for understanding tissue spatial organization. Current cell-type annotation methods often rely on extensive reference images and manual adjustments. In this work, we present a tool, the Robust Image-Based Cell Annotator (RIBCA), that enables accurate, automated, unbiased, and fine-grained cell-type annotation for images with a wide range of antibody panels without requiring additional model training or human intervention. Our tool has successfully annotated over 3 million cells, revealing the spatial organization of various cell types across more than 40 different human tissues. It is open source and features a modular design, allowing for easy extension to additional cell types.

The COVID-19 pandemic has resulted in millions of deaths and affected socioeconomic structure worldwide and the search for new antivirals and treatments are still ongoing. In the search for new drug targets and to increase our understanding of the disease, we applied large-scale immunofluorescence profiling to explore host cell response to SARS-CoV-2 infection. Among the 602 host proteins studied in this host response profiling, changes in abundance and subcellular localization were observed for 97 proteins, with 45 proteins showing increased abundance and 10 reduced abundance. 20 proteins displayed changed localization upon infection and an additional 22 proteins displayed altered abundance and localization, together contributing to diverse reshuffling of the host cell protein landscape during infection. We then selected existing and approved small-molecule drugs (n = 123) against our identified host response proteins and identified one compound - elesclomol, that significantly reduced antiviral activity. Our study introduces a novel, targeted and systematic approach based on host protein profiling, to identify new targets for drug repurposing. The dataset of > 100,000 immunofluorescence images from this study are published as a resource available for further studies. Author summary: In this study we have evaluated a new approach for identifying drugs that could be used as antiviral drugs, in this case demonstrated for SARS CoV-2. By mining the literature for reported interactions between SARS CoV-2 viral components and host cell proteins, we identified a few hundred host proteins suggested to interact with the virus upon infection. To explore these viral-host interaction proteins further, we developed an image based assay using immunofluorescence and confocal microscopy to visualize the host proteins within infected and non infected cells. This was possible due to the proteome wide collection of antibodies generated within the Human Protein Atlas project, with the aim to systematically map the human proteome in cells and across tissues. The host proteins that altered their location or abundance level upon infection were regarded as putative targets for drug repurposing and we subsequently tested 123 drugs that were targeting a subset of these host proteins. Applying these drugs on two different cell types infected with SARS-CoV-2, revealed a non toxic antiviral effect for one compound that can be explored further as a treatment regimen for SARS-CoV-2 infection. The approach is novel since it combines a targeted approach for drug repurposing screening, giving insight into mechanism of action from start. As such it has the potential to accelerate drug repurposing or identification of targets for new drugs.

Primary cilia are critical organelles found on most human cells. Their dysfunction is linked to hereditary ciliopathies with a wide phenotypic spectrum. Despite their significance, the specific roles of cilia in different cell types remain poorly understood due to limitations in analyzing ciliary protein composition. We employed antibody-based spatial proteomics to expand the Human Protein Atlas to primary cilia. Our analysis identified the subciliary locations of 715 proteins across three cell lines, examining 128,156 individual cilia. We found that 69% of the ciliary proteome is cell-type specific, and 78% exhibited single-cilia heterogeneity. Our findings portray cilia as sensors tuning their proteome to effectively sense the environment and compute cellular responses. We reveal 91 cilia proteins and found a genetic candidate variant in CREB3 in one clinical case with features overlapping ciliopathy phenotypes. This open, spatial cilia atlas advances research on cilia and ciliopathies.

Scientific research demands sophisticated reasoning over multimodal data, a challenge especially prevalent in biology. Despite recent advances in multimodal large language models (MLLMs) for AI-assisted research, existing multimodal reasoning benchmarks only target up to college-level difficulty, while research-level benchmarks emphasize lower-level perception, falling short of the complex multimodal reasoning needed for scientific discovery. To bridge this gap, we introduce MicroVQA, a visual-question answering (VQA) benchmark designed to assess three reasoning capabilities vital in research workflows: expert image understanding, hypothesis generation, and experiment proposal. MicroVQA consists of 1,042 multiple-choice questions (MCQs) curated by biology experts across diverse microscopy modalities, ensuring VQA samples represent real scientific practice. In constructing the benchmark, we find that standard MCQ generation methods induce language shortcuts, motivating a new two-stage pipeline: an optimized LLM prompt structures question-answer pairs into MCQs; then, an agent-based 'RefineBot' updates them to remove shortcuts. Benchmarking on state-of-the-art MLLMs reveal a peak performance of 53%; models with smaller LLMs only slightly underperform top models, suggesting that language-based reasoning is less challenging than multimodal reasoning; and tuning with scientific articles enhances performance. Expert analysis of chain-of-thought responses shows that perception errors are the most frequent, followed by knowledge errors and then overgeneralization errors. These insights highlight the challenges in multimodal scientific reasoning, showing MicroVQA is a valuable resource advancing AI-driven biomedical research. MicroVQA is available here, project here.

Human cells consist of a complex hierarchy of components, many of which remain unexplored1,2. Here we construct a global map of human subcellular architecture through joint measurement of biophysical interactions and immunofluorescence images for over 5,100 proteins in U2OS osteosarcoma cells. Self-supervised multimodal data integration resolves 275 molecular assemblies spanning the range of 10-8 to 10-5 m, which we validate systematically using whole-cell size-exclusion chromatography and annotate using large language models3. We explore key applications in structural biology, yielding structures for 111 heterodimeric complexes and an expanded Rag-Ragulator assembly. The map assigns unexpected functions to 975 proteins, including roles for C18orf21 in RNA processing and DPP9 in interferon signalling, and identifies assemblies with multiple localizations or cell type specificity. It decodes paediatric cancer genomes4, identifying 21 recurrently mutated assemblies and implicating 102 validated new cancer proteins. The associated Cell Visualization Portal and Mapping Toolkit provide a reference platform for structural and functional cell biology.

Heart development relies on topologically orchestrated cellular transitions and interactions, many of which remain poorly characterized in humans. Here, we combined unbiased spatial and single-cell transcriptomics with imaging-based validation across postconceptional weeks 5.5 to 14 to uncover the molecular landscape of human early cardiogenesis. We present a high-resolution transcriptomic map of the developing human heart, revealing the spatial arrangements of 31 coarse-grained and 72 fine-grained cell states organized into distinct functional niches. Our findings illuminate key insights into the formation of the cardiac pacemaker-conduction system, heart valves and atrial septum, and uncover unexpected diversity among cardiac mesenchymal cells. We also trace the emergence of autonomic innervation and provide the first spatial account of chromaffin cells in the fetal heart. Our study, supported by an open-access spatially centric interactive viewer, offers a unique resource to explore the cellular and molecular blueprint of human heart development, offering links to genetic causes of heart disease.

Microscopy-based spatially resolved omic methods are transforming the life sciences. However, these methods rely on high numerical aperture objectives and cannot resolve crowded molecular targets, limiting the amount of extractable biological information. To overcome these limitations, here we develop Deconwolf, an open-source, user-friendly software for high-performance deconvolution of widefield fluorescence microscopy images, which efficiently runs on laptop computers. Deconwolf enables accurate quantification of crowded diffraction limited fluorescence dots in DNA and RNA fluorescence in situ hybridization images and allows robust detection of individual transcripts in tissue sections imaged with ×20 air objectives. Deconvolution of in situ spatial transcriptomics images with Deconwolf increased the number of transcripts identified more than threefold, while the application of Deconwolf to images obtained by fluorescence in situ sequencing of barcoded Oligopaint probes drastically improved chromosome tracing. Deconwolf greatly facilitates the use of deconvolution in many bioimaging applications.

Lysine deacetylase inhibitors (KDACis) are approved drugs for cutaneous T cell lymphoma (CTCL), peripheral T cell lymphoma (PTCL), and multiple myeloma, but many aspects of their cellular mechanism of action (MoA) and substantial toxicity are not well understood. To shed more light on how KDACis elicit cellular responses, we systematically measured dose-dependent changes in acetylation, phosphorylation, and protein expression in response to 21 clinical and pre-clinical KDACis. The resulting 862,000 dose-response curves revealed, for instance, limited cellular specificity of histone deacetylase (HDAC) 1, 2, 3, and 6 inhibitors; strong cross-talk between acetylation and phosphorylation pathways; localization of most drug-responsive acetylation sites to intrinsically disordered regions (IDRs); an underappreciated role of acetylation in protein structure; and a shift in EP300 protein abundance between the cytoplasm and the nucleus. This comprehensive dataset serves as a resource for the investigation of the molecular mechanisms underlying KDACi action in cells and can be interactively explored online in ProteomicsDB.