Cross-Propagative Graph Learning Reveals Spatial Tissue Domains in Multi-Modal Spatial Transcriptomics

Abstract

Spatial transcriptomics enables in situ characterization of tissue organization by jointly profiling gene expression profiles and spatial coordinates, with histological images as complementary contextual information. However, effectively integrating these heterogeneous modalities remains challenging due to differences in statistical properties and structural patterns. We propose st-Xprop, a cross-propagative graph network with dual-graph embedding coupling for spatial domain identification. st-Xprop constructs modality-specific graphs for gene expression and histological features, and performs alternating cross-modal propagation to explicitly model inter-modal heterogeneity while enabling complementary information exchange. Through dual-graph embedding coupling, the framework progressively learns a unified low-dimensional representation that integrates multimodal signals and preserves spatial coherence. Evaluations on multiple real spatial transcriptomics datasets demonstrate that st-Xprop consistently improves clustering accuracy and robustness, particularly in weak-signal or structurally complex regions, yielding spatial domains that are more stable and biologically meaningful.