DGAE: Dynamic Graph Convolutional Network for Multi-Slice Spatial Transcriptomics Alignment and Enhancement.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics (ST) analyzes gene expression within tissue context.
• Integrating multiple ST slices is challenging due to varying spatial distributions.
• Traditional graph convolutions struggle with dynamic spatial relationships.

Purpose of the Study:

• To develop a novel framework for multi-slice spatial transcriptomics data alignment and enhancement.
• To address limitations of fixed graph structures in capturing spatial variations across tissue slices.
• To improve the analysis of cell interactions, developmental processes, and disease progression.

Main Methods:

• Proposed DGAE (Dynamic Graph Autoencoder) framework based on Dynamic Graph Convolutional Neural Networks (DGCNN).
• DGAE_align module uses K-nearest neighbor (KNN) and r-radius for hybrid graph construction and spatial alignment.
• DGAE_recog module aggregates information from adjacent slices for data enhancement.

Main Results:

• DGAE significantly outperforms existing methods in multi-slice ST data alignment.
• DGAE demonstrates superior performance in ST data enhancement tasks.
• The framework shows adaptability and stability in spatial domain recognition, denoising, and disease research.

Conclusions:

• DGAE provides an effective solution for multi-slice spatial transcriptomics data integration.
• The framework offers wide applicability and scalability for various ST analysis tasks.
• DGAE enhances the understanding of biological processes through improved spatial data analysis.