Accurate reconstruction of spatial cell-type maps and characterization of domain-specific functions based on a gene-aware heterogeneous network.

STGnet achieves the best overall performance in spatial transcriptomic deconvolution and functional annotation compared to state-of-the-art methods.

• Why it matters: Accurate mapping of cell types and functions within spatial transcriptomic data is crucial for understanding tissue organization and disease mechanisms, but existing methods often produce fragmented maps and lack interpretability.
• What they did: The authors developed STGnet, a gene-aware heterogeneous graph attention network that uses a hybrid pseudospot generation strategy to integrate spatial adjacency, transcriptional similarity, and gene-spot associations in a unified model, with attention weights highlighting domain-specific genes.
• The result: STGnet accurately reconstructs spatial cell-type maps and characterizes domain-specific functional programs, including those associated with disease progression, enabling better insights into spatial disease mechanisms.