Frequency-Aware Dual-Graph Fusion for Spatial Domain Recognition

Spatial domain identification is essential to developing a more profound understanding of the tissue microenvironment in biological systems. The advancement of spatial transcriptomics (ST) technologies has enabled the comprehensive assessment of gene expression patterns while preserving the spatial context inherent to the tissue microenvironment. However, due to the high-dimensional and noise-prone nature of gene expression data, ST technology struggles to accurately distinguish and preserve the true biological heterogeneity signals within the data, which limits the precise interpretation of the tissue’s spatial context. To address this challenge, this paper proposes a frequency-aware dual-graph fusion-based network architecture (WTGFN), aimed at capturing the authentic spatial transcriptomics features. This framework innovatively introduces an “expression similarity global graph and spatial radius local graph” interaction processing and dynamic weight adjustment strategy, balancing the distant isomorphic microenvironments and local continuity, thereby obtaining feature representations that reflect the true biological relationships of the tissue microenvironment. Experimental results demonstrate that WTGFN performs exceptionally well across different datasets.