Spatial pattern enhanced cellular and tissue recognition for spatial transcriptomics

Spatially mapping the cellular positions and their microenvironments with spatial transcriptomics (ST) shows great potential to illustrate key factors and mechanisms driving complex tissue organizations. The spatial data require specialized handling with different statistical and inferential considerations. Here, we develop SPECTRUM (Spatial Pattern Enhanced Cellular and Tissue Recognition Unified Method), which combines inclusive prior known cell-type-specific markers and spatial weighting for cell-type identification and spatial community detection. Comprehensive benchmarks demonstrate the superior performance of SPECTRUM. Applying SPECTRUM on real ST datasets with various spatial patterns demonstrates its capability in correctly mapping region-specific cell types and functional spatial communities. With that, we uncovered that context-dependent communication supports the functional plasticity of cells in spatial communities in human limb development. In summary, SPECTRUM is a unified tool for ST data analysis that deepens our insights into spatial organization at molecular, cellular, and community levels.