MCGAE

MCGAE creates multi-view representations from gene expression and spatial adjacency matrices for spatial transcriptomics data analysis, integrating multimodal data including spatial coordinates and morphological information to advance spatial domain identification.

Key Features:

• Multi-view representation learning: Generates modality-specific spatial representations from gene expression and spatial adjacency matrices that are integrated into a unified embedding.
• Contrastive graph convolutional networks: Utilizes contrastive learning with graph convolutional network architectures for representation learning.
• Attention mechanisms: Incorporates attention mechanisms to weight and integrate features across modalities.
• Morphological image feature integration: Enriches the integration process by including morphological image features for enhanced multimodal data processing.
• Spatial domain detection: Identifies spatial domains with improved accuracy compared to existing methods.
• Data denoising: Reduces noise in spatial transcriptomics data as part of the analytical pipeline.
• Trajectory inference: Infers developmental or pseudotemporal trajectories from spatial transcriptomics data.
• 3D feature extraction: Extracts three-dimensional features from spatial transcriptomics datasets.

Scientific Applications:

• Tumor invasion analysis: Integrates histological and gene expression data to identify tumor invasion regions in colorectal cancer liver metastases.
• Cellular molecular regulation characterization: Characterizes cellular molecular regulation patterns within tumor microenvironments.
• Cancer research: Provides analytical capabilities for studying cancer progression and heterogeneity through spatial transcriptomics.
• Complex disease research: Extends spatial transcriptomics analysis to support research on complex diseases beyond cancer.
• Tissue microenvironment heterogeneity analysis: Reveals heterogeneity of tissue microenvironments through integrated multimodal spatial transcriptomics.

Methodology:

MCGAE employs modular modeling with contrastive graph convolutional networks and attention mechanisms to generate modality-specific spatial representations from gene expression and spatial adjacency matrices, which are integrated into a unified embedding enriched with morphological image features.