M-STGCN: A Position-Aware Multimodal Graph Convolutional Framework for Joint Spatial Domain Identification and Gene...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatially transcriptomics (ST) offers insights into tissue microenvironments but requires multimodal data integration to address biases.
• Decoding spatial heterogeneity necessitates unifying gene expression, spatial positions, and histopathological images.

Purpose of the Study:

• To develop M-STGCN, a multimodal unsupervised framework for integrating diverse ST data.
• To enhance spatial domain identification accuracy by refining gene expression and image features using spatial coordinates.

Main Methods:

• Constructed a spatial weight matrix from spatial coordinates for position-aware feature enhancement.
• Employed graph fusion to integrate refined gene expression and image features.
• Validated the framework on human brain and breast cancer datasets.

Main Results:

• M-STGCN significantly improved spatial domain identification accuracy.
• Ablation studies confirmed the importance of position-aware and image modality integration.
• The framework demonstrated robust performance even with limited data modalities (gene expression and spatial coordinates only).
• Identified significant spatial domain marker genes and potential prognostic biomarkers for breast cancer.
• Revealed equal contributions of image and spatial information in breast cancer analysis.

Conclusions:

• M-STGCN provides an unbiased and scalable method for multimodal data integration in ST.
• The framework effectively denoises ST profiles, facilitating the discovery of spatial biomarkers.
• Highlights the critical role of histopathological images in understanding tissue heterogeneity.