STELLA: a spatial transcriptomics framework for microenvironment decoding using dynamic graph neural networks.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics enables gene expression analysis with spatial context.
• Accurate identification of spatial domains and intercellular communication remains challenging.

Purpose of the Study:

• To introduce STELLA, a novel framework for analyzing spatial transcriptomics data.
• To enhance the identification of spatial domains and intercellular communication networks.

Main Methods:

• STELLA integrates dynamic graph neural networks and self-supervised learning.
• It employs a complementary space-feature dual-graph structure and dynamic adjacency matrix learning.
• Dual-channel graph convolutional networks and multi-head attention process spatial and expression information.

Main Results:

• STELLA accurately identified the tumor-muscle interface in zebrafish melanoma, linking MT-CO1 pathway to invasion.
• It detected immune aggregation in intestinal tissues and implicated cyclosporin A signaling.
• STELLA revealed bidirectional regulatory networks (PERIOSTIN and MHC-II) in mouse striatum and analyzed axolotl brain regeneration processes.

Conclusions:

• STELLA provides a robust framework for spatial transcriptomics data analysis.
• The framework accurately identifies biological spatial domains and communication networks.
• STELLA offers novel insights into various biological processes, including cancer invasion and tissue regeneration.