Graspot: a graph attention network for spatial transcriptomics data integration with optimal transport.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) technologies provide gene expression data with spatial coordinates, enabling 3D tissue reconstruction.
• Integrating multiple ST datasets is crucial for comprehensive analysis but faces challenges in removing batch effects while preserving biological variations.

Purpose of the Study:

• To introduce Graspot, a novel computational method for integrating spatial transcriptomics data.
• To address the challenge of batch effect removal in ST data integration while maintaining biological structure.

Main Methods:

• Graspot utilizes a graph attention network combined with unbalanced optimal transport.
• It integrates gene expression and spatial information to align common structures across multiple ST datasets.
• The method embeds datasets into a unified latent space for partial alignment of spots from different slices.

Main Results:

• Graspot demonstrates superior performance compared to existing methods on four real ST datasets.
• It excels in ST data integration tasks, including those requiring partial alignment.
• The method successfully integrates multiple ST slices, guides coordinate alignment, and reconstructs human heart development.

Conclusions:

• Graspot offers an effective solution for spatial transcriptomics data integration.
• The method accurately removes batch effects and preserves biological variations, facilitating 3D tissue reconstruction and developmental studies.