DSTG

DSTG deconvolves spatial transcriptomics (ST) spot-level gene expression using graph-based convolutional networks (GCNs) to estimate cell-type compositions and reveal spatial cellular architecture.

Key Features:

• Graph-based convolutional networks (GCNs): Applies GCNs to model spatial relationships among spots and integrate gene expression for deconvolution.
• Spot-level deconvolution: Disentangles mixed gene expression from spots containing multiple cell types to estimate cell-type proportions.
• Cell composition recovery: Recovers cell constitutions within individual spatial spots to enable analysis of cellular heterogeneity.
• Spatial segmentation: Enables high-level segmentation to reveal the spatial architecture of cellular heterogeneity within tissues.
• Validation on synthetic data: Demonstrates robust performance on synthetic spatial data generated from diverse protocols.
• Empirical applications: Applied to biological samples including mouse cortex layers, hippocampus slices, and pancreatic tumor tissues.

Scientific Applications:

• Cell-state and subpopulation identification: Identifies cell states and subpopulations based on their spatial localization within tissue sections.
• Mapping tissue cytoarchitecture: Reconstructs spatial distributions of cell types to study tissue organization and function.
• Analysis of complex biological samples: Characterizes spatial compositions in mouse cortex layers, hippocampus slices, and pancreatic tumor tissues.
• Benchmarking deconvolution methods: Provides validation using synthetic spatial data from diverse protocols for method comparison.

Methodology:

Employs graph-based convolutional networks (GCNs) to model spatial relations and deconvolve spot-level gene expression, recovering cell-type compositions and enabling spatial segmentation.