smiMAD: Adaptive multi-view graph learning for paired spatial omics integration.

smiMAD achieves competitive clustering accuracy and spatial coherence in 17 simulated and 15 real paired spatial omics datasets.

• Why it matters: Integrating spatial transcriptomic and epigenomic or proteomic data is challenging due to differences in feature distributions, sparsity, and spatial dependencies, limiting the preservation of spatial organization and modality-specific information.
• What they did: smiMAD models spatial-neighborhood and molecular-feature graphs with view-specific and common-information branches, using an adaptive gate and alignment-decorrelation objectives, along with a dynamic spatial-consistency strategy across multiple datasets.
• The result: The method's integration performance and molecular marker patterns support accurate spatial region inference, enabling improved multimodal spatial omics analysis.