Unraveling spatial domain characterization in spatially resolved transcriptomics with robust graph contrastive...

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Spatial transcriptomics enables gene expression and distribution analysis in tissues.
• Understanding cellular interactions is key for tissue heterogeneity, regeneration, and disease.
• Current spatial clustering methods often underutilize spatial information, leading to inaccurate domain identification.

Purpose of the Study:

• To develop an advanced deep graph contrastive clustering framework, stDGCC.
• To accurately identify spatial domains by effectively modeling spatial and gene expression data.
• To improve the analysis of spatial transcriptomics data for biological insights.

Main Methods:

• Developed stDGCC, a deep graph contrastive clustering framework.
• Employed a spatially informed graph node embedding model to retain topological information.
• Utilized self-supervised contrastive learning for informative data characterization.
• Jointly optimized contrastive learning, reconstruction, and KL divergence losses for end-to-end learning.

Main Results:

• stDGCC accurately uncovers underlying spatial domains in spatial transcriptomics data.
• The framework effectively preserves topological structure and optimizes feature learning.
• Validated on diverse datasets across different platforms and spatial resolutions.
• Demonstrated superior performance compared to state-of-the-art clustering methods in identifying cellular structures.

Conclusions:

• stDGCC offers a superior approach for spatial domain identification in transcriptomics.
• The method accurately identifies cellular-level biological structures.
• Provides a powerful tool for advancing research in tissue heterogeneity, regeneration, and disease.