Dual-contrastive learning for spatial domain identification in spatial transcriptomics with STAMGC.

STAMGC achieves improved spatial domain identification in spatial transcriptomics, outperforming existing methods across multiple datasets with enhanced detail and noise reduction.

• Why it matters: Accurate spatial domain detection is crucial for understanding organism development, but current models struggle to balance local detail preservation and noise suppression.
• What they did: The authors developed STAMGC, a dual-contrastive learning framework based on graph convolutional networks, incorporating Gaussian smoothing for region-level contrastive learning.
• The result: STAMGC reveals finer structures in the mouse brain and uncovers new insights in human breast cancer, enabling more precise spatial analysis and biological discovery.