SpaInGNN: Enhanced clustering and integration of spatial transcriptomics based on refined graph neural networks

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) enables gene expression analysis within tissue context.
• Identifying single-cell level spatial domains remains a challenge for biological process elucidation.

Purpose of the Study:

• To develop a novel graph neural network (GNN) framework, SpaInGNN, for accurate spatial domain delineation.
• To integrate spatial location, histological, and gene expression data for enhanced tissue microenvironment characterization.

Main Methods:

• SpaInGNN refines spatial graphs using tissue image and Euclidean distances after gene expression pre-clustering.
• A self-supervised GNN embeds the refined graph, minimizing self-reconfiguration loss.
• Application across multiple tissue slices mitigates batch effects.

Main Results:

• SpaInGNN demonstrates significant improvements in spatial domain recognition accuracy.
• The method provides a more faithful representation of tissue organization.
• Successful application in mouse olfactory bulb and human lateral prefrontal cortex samples.

Conclusions:

• SpaInGNN effectively delineates spatial domains at the single-cell level.
• The framework offers a robust approach for analyzing spatial transcriptomics data.
• This technology advances the understanding of complex tissue microenvironments.