PathCLAST: pathway-augmented contrastive learning with attention for interpretable spatial transcriptomics.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Understanding molecular programs in tissues is crucial for tumor evolution and microenvironment studies.
• Current spatial transcriptomics methods lack biological interpretability and pathway topology integration.

Purpose of the Study:

• To develop a novel framework for interpretable spatial transcriptomics analysis.
• To integrate gene expression, histopathology, and pathway structures for accurate domain identification.

Main Methods:

• Introduced PathCLAST (Pathway-augmented Contrastive Learning with Attention for interpretable Spatial Transcriptomics).
• Utilized bi-modal contrastive learning to integrate gene expression, histopathology images, and pathway graphs.
• Employed pathway embedding for biology-informed dimensionality reduction.

Main Results:

• Achieved state-of-the-art spatial domain identification on multiple public datasets.
• Provided pathway-level attention scores for mechanistic interpretation.
• Uncovered domain-specific pathways, signaling activities, intra-domain heterogeneity, and inter-domain crosstalk.

Conclusions:

• PathCLAST offers accurate and biologically interpretable spatial domain delineation.
• The framework provides fine-grained insights into tumor progression and tissue architecture.
• PathCLAST enhances translational impact in cancer research and spatial biology.