HarveST uses a heterogeneous graph learning framework to reveal spatial transcriptomics patterns.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics allows in situ gene expression profiling.
• Accurate spatial domain identification and marker gene detection are critical but challenging.

Purpose of the Study:

• To introduce HarveST, a novel heterogeneous graph-based framework for spatial transcriptomics analysis.
• To improve the identification of spatial domains and their marker genes.
• To enable joint analysis across multiple spatial transcriptomics sections.

Main Methods:

• HarveST integrates spatial, transcriptomic, and gene-gene interaction data using a unified computational model.
• It employs dual learning strategies: self-supervised learning for feature extraction and partially supervised refinement for domain delineation.
• A Random Walk with Restart algorithm identifies spatial domain-marker spatially variable genes (SVGs).

Main Results:

• HarveST demonstrated superior performance in detecting biologically meaningful spatial domains and marker genes across human cortical tissue, mouse olfactory bulb, and tumor microenvironments.
• The framework successfully supports joint analysis across consecutive spatial transcriptomics sections for consistent domain reconstruction.
• HarveST captures spatial topology and molecular relationships within a single graph-theoretical framework.

Conclusions:

• HarveST advances spatial transcriptomics analysis beyond conventional clustering methods.
• It offers deeper insights into tissue architecture and cellular interactions in both normal and pathological contexts.
• The framework provides a powerful tool for precise spatial domain and marker gene identification.