Spatially Informed Graph Structure Learning Extracts Insights from Spatial Transcriptomics.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics (ST) data analysis benefits from cell graph embeddings.
• Existing methods using spatial proximity for graph structures inadequately capture diverse cell-cell interactions (CCIs).

Purpose of the Study:

• Introduce STAGUE, a novel framework for concurrent cell graph structure learning and low-dimensional embedding from ST data.
• Enhance the accuracy of spatial clustering and the discovery of novel CCIs.

Main Methods:

• STAGUE utilizes graph structure learning to refine cell graph adjacency matrices.
• Learns graph views for effective contrastive learning.
• Generates low-dimensional embeddings and cell graph structures.

Main Results:

• STAGUE outperforms 15 comparison methods in clustering performance across 86 ST datasets.
• Delineates heterogeneity in human breast cancer tissues, identifying epithelial-to-mesenchymal transition and PI3K/AKT signaling.
• Identifies CCIs with improved biological alignment and reveals regulatory genes in signaling pathways.

Conclusions:

• STAGUE provides a robust framework for analyzing ST data, improving spatial clustering and CCI discovery.
• The method offers significant advancements in understanding tissue heterogeneity and biological processes through gene regulatory network analysis.