SpaGIC: graph-informed clustering in spatial transcriptomics via self-supervised contrastive learning.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics offers gene expression data with spatial context, crucial for understanding tissue heterogeneity.
• Effective integration of gene and spatial data is vital for accurate spatial domain identification.
• Existing methods often underutilize local neighborhood information within spatial data.

Purpose of the Study:

• To introduce SpaGIC, a novel graph-based deep learning framework for spatial transcriptomics analysis.
• To improve the exploitation of local neighborhood details in spatial information.
• To enhance tasks like spatial domain identification, data denoising, visualization, and trajectory inference.

Main Methods:

• SpaGIC utilizes graph convolutional networks and self-supervised contrastive learning.
• It learns latent embeddings by maximizing mutual information of graph structures.
• Minimizes embedding distance between spatially adjacent spots to leverage neighborhood information.

Main Results:

• SpaGIC demonstrated superior performance across seven diverse spatial transcriptomics datasets.
• Outperformed state-of-the-art methods in spatial domain identification, denoising, visualization, and trajectory inference.
• Successfully performed joint analyses of multiple tissue slices, highlighting its versatility.

Conclusions:

• SpaGIC provides a robust and versatile framework for spatial transcriptomics research.
• Its graph-based deep learning approach effectively captures local spatial context.
• Offers significant advancements in analyzing gene expression with spatial information.