Graph convolutional networks for inferring cell-cell communication from spatial transcriptomics data.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell spatial transcriptomics enables gene expression analysis within tissue context.
• Inferring cell-cell communication (CCC) is crucial for understanding tissue function.
• Existing CCC inference methods struggle with simplistic spatial context representations.

Purpose of the Study:

• To develop a more accurate method for inferring CCC from spatial transcriptomics data.
• To address the limitations of existing models that use simplistic spatial representations.
• To improve the validity of CCC inference by enhancing spatial context modeling.

Main Methods:

• Utilizing a graph convolutional network (GCN) as a spatially informed model, representing cells as nodes and spatial proximity as edges.
• Comparing a GCN-based model with existing approaches on semi-synthetic datasets.
• Validating the method on MERFISH and Xenium mouse brain tissue datasets.

Main Results:

• The GCN-based approach avoids spurious CCC inferences often produced by simplistic neighborhood feature models.
• The method successfully identifies genes with known spatial variation in mouse brain tissue.
• Demonstrated improved accuracy and reliability in CCC inference compared to existing methods.

Conclusions:

• Graph convolutional networks offer a powerful and expressive approach for modeling spatial context in transcriptomics.
• The proposed GCN-based method enhances the accuracy and validity of cell-cell communication inference.
• This work provides a robust tool for analyzing intercellular interactions in complex biological tissues.