SpatioCell: A Deep Learning Algorithm for High-resolution Single-cell Mapping through Deep Integration of Histology Image and Sequencing Data

Spatial transcriptomics (ST) is a powerful assay to capture gene expression in tissue context. However, due to the limitation of resolution, most existing ST datasets remain at multicellular resolution which hinders comprehensive understanding of the spatial organization. We propose SpatioCell, a computational algorithm to automatically extract both cell type and expression information at single-cell resolution from ST data, through a morpho-transcriptomic spatial reconstruction framework solved via dynamic programming, integrating morphological and transcriptomic information. This framework enables deterministic single-cell spatial reconstruction, assigning cell identities to precise locations rather than inferring spot-level cell-type composition and expression, and further uncovers overlooked microenvironment features while correcting deconvolution errors. Using the ST data from triple negative breast cancer as an example, SpatioCell reveals the significance of the distance between cancer-associated fibroblast (CAF) and tumor or immune cells for disease progression. The establishment of SpatioCell will broaden the biomedical applications of ST and facilitate investigations of single-cell spatial organization.