Heterogeneous graph contrastive learning for integration and alignment of spatial transcriptomics data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) provides gene expression and spatial data from tissue slices.
• Independent analysis of ST slices misses shared features, limiting biological discovery.
• Integrating multislice ST data is crucial for comprehensive understanding.

Purpose of the Study:

• To develop a novel framework, GRASS, for integrating and aligning multislice spatial transcriptomics data.
• To leverage deep graph representation learning for enhanced analysis of ST data.
• To improve biological insights by analyzing shared and unique features across multiple tissue slices.

Main Methods:

• GRASS utilizes a deep graph representation learning framework with two modules: GRASS_Integration and GRASS_Alignment.
• GRASS_Integration employs a heterogeneous graph, contrastive learning, and multi-expert collaboration for data integration.
• GRASS_Alignment uses a dual-perception similarity metric for spot-level alignment and downstream tasks like 3D reconstruction.

Main Results:

• GRASS demonstrated superior performance in integrating and aligning multislice ST data across seven datasets from five platforms.
• The framework consistently outperformed eight state-of-the-art methods in benchmark evaluations.
• GRASS effectively captures both shared and unique information for comprehensive multislice analysis.

Conclusions:

• GRASS provides an effective solution for the joint analysis of multislice spatial transcriptomics data.
• The framework enhances biological insights by integrating and aligning data across multiple tissue slices.
• GRASS represents a significant advancement in computational tools for spatial transcriptomics research.