Decoding Spatial Heterogeneity and Multi-Omics Regulation with Hierarchical Graph Learning

Recent advances in spatial multi-omics technologies have enabled the simultaneous profiling of multiple molecular layers within the same tissue slice, providing unprecedented opportunities to investigate tissue spatial organization. However, most existing computational methods identify spatial domains in a purely data-driven manner, rarely uncovering the interpretable multi-omics regulatory mechanisms that underlie spatial heterogeneity. Here, we introduce SMOReg, a spatial multi-omics hierarchical graph learning framework that embeds cross-omics regulatory patterns into spot representations via cross-graph matching, thereby yielding fine-grained spatial domains with enhanced biological interpretability. Evaluated on paired spatial transcriptomic and proteomic datasets from diverse tissues, SMOReg consistently outperformed existing methods across multiple metrics. Notably, SMOReg uncovered fine-grained spatial domains characterized by biologically interpretable multi-layer regulatory signatures that competing approaches overlooked. It successfully distinguished the germinal center light zone and dark zone in human tonsil, which are histologically well-established yet spatially ambiguous compartments, and revealed their specialized regulatory pathways driving B-cell proliferation and affinity maturation. SMOReg thus provides a powerful framework for deciphering the interplay between spatial domain heterogeneity and underlying multi-omics regulation, significantly advancing the interpretability of spatial domains in complex tissues.