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

STAMGC, a dual-contrastive learning framework based on graph convolutional networks, outperforms existing methods in spatial domain identification in spatial transcriptomics data.

• Why it matters: Accurately identifying spatial domains in spatial transcriptomics remains difficult due to the challenge of balancing local detail preservation and noise reduction, which limits understanding of organism development and disease.
• What they did: The authors developed STAMGC, incorporating regional and topological contrastive learning, and applied Gaussian smoothing to process spatial transcriptomics data, enabling more precise spatial domain detection across multiple datasets.
• The result: STAMGC demonstrates superior performance in identifying finer structures in mouse brain and uncovers new insights in human breast cancer research, advancing the potential for detailed spatial analysis in biomedical studies.