Hypergraph-driven spatial multimodal fusion for precise domain delineation and tumor microenvironment decoding.

Area of Science:

• Oncology
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics offers insights into tumor microenvironment complexity.
• Accurate spatial domain delineation is crucial but challenging for existing methods.
• Current techniques struggle with single-view data and complex spot relationships.

Purpose of the Study:

• To develop a novel tool, HAST, for precise spatial domain delineation and tumor microenvironment decoding.
• To overcome limitations of existing methods by integrating multimodal spatial data.
• To enhance understanding of spatial heterogeneity and cellular interactions within tumors.

Main Methods:

• HAST utilizes a hypergraph-driven approach to fuse gene expression, spatial coordinates, and histological features.
• It constructs local hypergraphs modeling many-to-many spatial relationships, aggregated into a global hypergraph.
• A hypergraph convolutional network with self-supervised contrastive learning learns fused representations.

Main Results:

• HAST accurately delineates spatial domains and identifies domain-associated genes across multiple datasets.
• Performance benchmarks show HAST outperforms state-of-the-art methods.
• Functional enrichment analyses reveal biologically relevant pathways and novel tumor microenvironment insights.

Conclusions:

• HAST provides a robust framework for decoding tumor spatial complexity using multimodal data.
• The tool enables precise spatial omics analyses in cancer research.
• HAST advances the study of tumor microenvironments and cellular interactions.