This study introduces TopACT, a multiscale method for automatically classifying cell types in subcellular spatial transcriptomics data. The method leverages both transcriptomic information and spatial context to identify cell types at a subcellular level. It is applied to both targeted and whole-transcriptome spatial platforms, improving cell classification and morphology in human kidney tissue and identifying sparse immune cells in mouse kidney data 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 using immunofluorescence. The framework generalizes to new platforms, providing a comprehensive pipeline bridging different levels of biological organization from genes to tissues. A key challenge in spatial transcriptomics is inferring information at the level of single cells. While recent technologies enable whole-transcriptome measurement at subcellular resolution, new computational methods are needed to infer single-cell information. TopACT provides mathematical tools to fill this gap without prior knowledge of cell boundaries. The method uses a two-dimensional point cloud and dynamically scaled local neighborhoods to classify each spot's cell type. It aggregates gene expression information across scales without prior knowledge of cell boundaries, enabling the identification of rare and sparsely distributed cells. TopACT was benchmarked against fixed-window decomposition on synthetic data and demonstrated superior accuracy in identifying rare cell types. It was also applied to mouse brain and kidney data, identifying previously undetected immune cells and improving the resolution of cell morphology. The method was validated using immunofluorescence and showed increased immune cell activity in treated samples, consistent with lupus-like immune infiltration. MPH landscapes were used to quantify the spatial organization of immune cells, predicting a ring structure in treated glomerular immune cells, which was verified with immunofluorescence. TopACT was also applied to human kidney data, identifying podocytes and other cell types with high accuracy. The method was shown to improve the resolution of cell morphology and enable the identification of rare cell types. It was further validated using immunofluorescence and showed increased immune cell activity in treated samples, consistent with lupus-like immune infiltration. The method was also applied to mouse kidney data, identifying immune cells and showing increased immune cell activity in treated samples, consistent with lupus-like immune infiltration. The study highlights the potential of TopACT as a powerful tool for spatial transcriptomics, enabling the identification of rare and sparsely distributed cells and providing insights into tissue organization.This study introduces TopACT, a multiscale method for automatically classifying cell types in subcellular spatial transcriptomics data. The method leverages both transcriptomic information and spatial context to identify cell types at a subcellular level. It is applied to both targeted and whole-transcriptome spatial platforms, improving cell classification and morphology in human kidney tissue and identifying sparse immune cells in mouse kidney data 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 using immunofluorescence. The framework generalizes to new platforms, providing a comprehensive pipeline bridging different levels of biological organization from genes to tissues. A key challenge in spatial transcriptomics is inferring information at the level of single cells. While recent technologies enable whole-transcriptome measurement at subcellular resolution, new computational methods are needed to infer single-cell information. TopACT provides mathematical tools to fill this gap without prior knowledge of cell boundaries. The method uses a two-dimensional point cloud and dynamically scaled local neighborhoods to classify each spot's cell type. It aggregates gene expression information across scales without prior knowledge of cell boundaries, enabling the identification of rare and sparsely distributed cells. TopACT was benchmarked against fixed-window decomposition on synthetic data and demonstrated superior accuracy in identifying rare cell types. It was also applied to mouse brain and kidney data, identifying previously undetected immune cells and improving the resolution of cell morphology. The method was validated using immunofluorescence and showed increased immune cell activity in treated samples, consistent with lupus-like immune infiltration. MPH landscapes were used to quantify the spatial organization of immune cells, predicting a ring structure in treated glomerular immune cells, which was verified with immunofluorescence. TopACT was also applied to human kidney data, identifying podocytes and other cell types with high accuracy. The method was shown to improve the resolution of cell morphology and enable the identification of rare cell types. It was further validated using immunofluorescence and showed increased immune cell activity in treated samples, consistent with lupus-like immune infiltration. The method was also applied to mouse kidney data, identifying immune cells and showing increased immune cell activity in treated samples, consistent with lupus-like immune infiltration. The study highlights the potential of TopACT as a powerful tool for spatial transcriptomics, enabling the identification of rare and sparsely distributed cells and providing insights into tissue organization.