Spatial cell-cell communication inference based on cell-spot-ligand-receptor heterogeneous graphs

Cell-cell communication is essential for tissue homeostasis and disease-associated microenvironment remodeling. Spatial transcriptomics enables the study of communication events in situ, but single-cell-resolution inference from spot-based data remains challenging because of uncertain cell-to-spot mapping, limited use of local spatial information, and false positives from ligand-receptor co-expression. Existing methods increasingly emphasize spatial proximity in cell-cell communication inference, while integrating biologically informative signals remains important for improving the interpretability of candidate communication events. Here, we present SpaHCC, a heterogeneous graph learning framework that integrates single-cell transcriptomics, spatial transcriptomics, and prior ligand-receptor knowledge for spatial cell-cell communication inference. By jointly leveraging molecular features, spatial neighborhood information, and cellular functional states, SpaHCC prioritizes candidate communication events and supports their biological contextualization through pathway, transcription factor, and receptor-side response analyses. Benchmark and spatial consistency analyses showed that SpaHCC recovered spatially coherent candidate communication signals. In Alzheimer's disease data, SpaHCC identified recurrent excitatory neuron-associated candidate patterns, including NRXN- and FLRT2-related modules associated with synaptic remodeling and inflammation-related transcriptional programs. Cross-patient analyses showed recurrence of selected AD-associated candidate patterns, while representative cSCC and PDAC analyses illustrated the spatial interpretability of the framework in additional disease contexts. Overall, SpaHCC provides an effective and interpretable framework for spatially resolved cell-cell communication analysis in complex diseases.