Cross-domain information fusion for enhanced immune cell population delineation in single-cell spatial-omics data 2975

Abstract Description Understanding the immune microenvironment requires precise delineation of immune cell populations and their spatial interactions within tissues. Advanced spatial-omics technologies enable simultaneous analysis of cellular biomarkers, spatial contexts, and tissue architecture. However, current analytical tools fall short in integrating this multidimensional data, limiting insights into immune cell functionality and heterogeneity. To address this, we present CellSNAP (Cell Spatio- and Neighborhood-informed Annotation and Patterning), a self-supervised computational framework that integrates molecular expression, spatial neighborhood contexts, and local histological features using graph neural networks. CellSNAP enables the de novo discovery of fine-grained immune cell subpopulations, overcoming the constraints of predefined cell atlases. Applying CellSNAP to spatial-omics datasets across various tissues, including spleen and lymphoma, revealed immune subpopulations with distinct spatial distributions and functional states, such as tumor-associated macrophages and spatially stratified B and T cells. These findings underscore the utility of integrating spatial and molecular data to advance immunological research, offering a robust tool for unraveling immune cell dynamics in health and disease. Funding Sources 1P01AI177687-01, 75N93019C00071, DP2AI171139, R01AI149672, U24CA224331, NSF 2345215, NSF 2245575. Topic Categories Technological Innovations in Immunology (TECH)