Graph deep learning enabled spatial domains identification for spatial transcriptomics.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatially resolved transcriptomics (ST) technologies are crucial for understanding organ function and tissue microenvironments.
• Accurate spatial domain identification is essential for analyzing genome heterogeneity and cellular interactions within tissues.

Purpose of the Study:

• To develop a novel graph deep learning (GDL) based spatial clustering approach for accurate spatial domain identification in ST data.
• To improve the understanding of tissue microenvironments and cellular interactions through advanced computational methods.

Main Methods:

• A graph deep learning (GDL) framework was constructed, integrating gene expression profiles and spatial positions from ST data.
• A deep graph infomax module with a residual gated graph convolutional neural network was employed for feature extraction.
• A Bayesian Gaussian mixture model was utilized for latent embeddings to generate distinct spatial domains.

Main Results:

• The proposed GDL-based method demonstrated superior performance in spatial domain identification compared to state-of-the-art GDL techniques.
• Experimental results on multiple ST datasets validated the effectiveness and accuracy of the developed approach.
• The method successfully delineated spatial domains, providing insights into tissue architecture and cellular heterogeneity.

Conclusions:

• The developed GDL-based spatial clustering approach offers a powerful tool for advancing spatial transcriptomics analysis.
• This method enhances the ability to identify spatial domains, contributing to a deeper understanding of biological systems.
• The findings highlight the potential of GDL in uncovering complex spatial patterns within biological tissues.