Cells exhibiting similar gene regulatory networks tend to spatially aggregate into distinct cell types or functional domains to complete cellular function and form tissue architecture. However, most investigations in spatially resolved transcriptomics data focus only on identifying single-cell co-expression, neglecting spatial constraints and influences of proximal cells. To address these limitations, we introduce SpaGRN, a statistical framework that integrates spatially aware regulatory relationships and extracellular signaling path information. Specifically, it enables the prediction of cell type- or functional domain-specific, spatially aware, causal, and dynamic intracellular regulatory networks. Our results show that SpaGRN outperforms other inference tools regarding the precision of regulon identification through both synthetic and real datasets. Furthermore, the utility and versatility of SpaGRN have been demonstrated by applying it to various platforms including Stereo-seq, STARmap, MERFISH, CosMx, Slide-seq, and 10x Visium, complex cancerous samples, and 3D datasets of developing Drosophila embryos and larvae. In these applications, SpaGRN not only identifies receptor-involved spatial regulons, but also extends our understanding of the underlying regulatory mechanisms associated with organogenesis and inflammation.

bioinformatics,spatial transcriptomics,Gene regulatory network,algorithm