SCmapST enables single-cell spatial reconstruction by seeded matches-driven integration of single-cell and spatial transcriptomics

Single cell (SC) RNA sequencing and spatial transcriptomics (ST) exhibit complementary limitations: they struggle to balance the cellular granularity inherent in SC data and the spatial coordinate unique to ST data. Here, we present SCmapST, a deep learning framework that leverages a novel seeded matches strategy to map single cells in SC data to spatial locations in ST data. SCmapST surpasses existing methods by consistently and robustly reconstructing global tissue topology while preserving local tissue heterogeneity. SCmapST successfully deciphers spatial co-localization patterns, resolves location-dependent neuronal heterogeneity, and delineates cell-type distributions within both discrete tumor domains and continuous liver zonation.