The paper introduces a multiscale approach called TopACT for automatically classifying cell types at the subcellular level in spatial transcriptomics data. Spatial transcriptomics measures gene expression at millions of locations within a tissue, but it often faces trade-offs between transcriptome depth, spatial resolution, and sample size. While image-based segmentation has improved this field, it is limited by imaging quality and tissue heterogeneity. In contrast, array-based technologies can measure the entire transcriptome at subcellular resolution across large samples. However, there are no established methods to directly leverage this information for individual cell annotation. TopACT integrates transcriptomic information and spatial context to classify cell types at the subcellular level. It uses a dynamically scaled local neighborhood to classify each spot independently, improving cell classification and morphology for human kidney tissue and identifying sparse renal immune cells in mouse models without relying on image data. By integrating these predictions into a topological pipeline based on multiparameter persistent homology (MPH), the method identifies spatial relationships characteristic of a mouse model of lupus nephritis, which is validated experimentally by immunofluorescence. The proposed framework generalizes to new platforms, providing a comprehensive pipeline that bridges different levels of biological organization from genes to tissues. The method is benchmarked on synthetic data and applied to real datasets, demonstrating its ability to resolve fine structural details and identify rare cells. TopACT is particularly effective in detecting small and sparse cells, improving cellular morphology and cell annotation in imaging-based spatial transcriptomic data. The method also enables the systematic quantification of immune cell spatial organization in kidney diseases, such as lupus nephritis, by inferring the spatial arrangement of immune cells using MPH landscapes. In conclusion, TopACT is a powerful and flexible mathematical methodology that resolves cell type information at subcellular resolution and accurately localizes elusive sparse cells. It can be applied to both imaging-based and array-based spatial transcriptomics data, offering a robust approach to expression aggregation in the absence of cell boundaries.The paper introduces a multiscale approach called TopACT for automatically classifying cell types at the subcellular level in spatial transcriptomics data. Spatial transcriptomics measures gene expression at millions of locations within a tissue, but it often faces trade-offs between transcriptome depth, spatial resolution, and sample size. While image-based segmentation has improved this field, it is limited by imaging quality and tissue heterogeneity. In contrast, array-based technologies can measure the entire transcriptome at subcellular resolution across large samples. However, there are no established methods to directly leverage this information for individual cell annotation. TopACT integrates transcriptomic information and spatial context to classify cell types at the subcellular level. It uses a dynamically scaled local neighborhood to classify each spot independently, improving cell classification and morphology for human kidney tissue and identifying sparse renal immune cells in mouse models without relying on image data. By integrating these predictions into a topological pipeline based on multiparameter persistent homology (MPH), the method identifies spatial relationships characteristic of a mouse model of lupus nephritis, which is validated experimentally by immunofluorescence. The proposed framework generalizes to new platforms, providing a comprehensive pipeline that bridges different levels of biological organization from genes to tissues. The method is benchmarked on synthetic data and applied to real datasets, demonstrating its ability to resolve fine structural details and identify rare cells. TopACT is particularly effective in detecting small and sparse cells, improving cellular morphology and cell annotation in imaging-based spatial transcriptomic data. The method also enables the systematic quantification of immune cell spatial organization in kidney diseases, such as lupus nephritis, by inferring the spatial arrangement of immune cells using MPH landscapes. In conclusion, TopACT is a powerful and flexible mathematical methodology that resolves cell type information at subcellular resolution and accurately localizes elusive sparse cells. It can be applied to both imaging-based and array-based spatial transcriptomics data, offering a robust approach to expression aggregation in the absence of cell boundaries.