SpaGraphCCI: Spatial cell-cell communication inference through GAT-based co-convolutional feature integration.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatially resolved transcriptomics (SRT) offers spatial context for cell-cell interactions (CCIs).
• Integrating multimodal SRT data for accurate CCI inference remains a challenge.
• Existing methods struggle to leverage both gene expression and imaging data effectively.

Purpose of the Study:

• To develop a novel deep learning method, SpaGraphCCI, for integrating multimodal SRT data.
• To enhance the inference of spatial cell-cell interactions (CCIs) by combining gene expression and spatial information.
• To provide a robust tool for deciphering CCIs in various biological contexts.

Main Methods:

• SpaGraphCCI employs co-convolution to integrate features from different SRT modalities.
• Gene expression and image features are projected into a unified low-dimensional space.
• A deep learning framework is utilized for robust CCI inference.

Main Results:

• SpaGraphCCI achieved high performance on single-cell (AUC 0.860-0.907) and spot resolution (AUC 0.880-0.965) datasets.
• Outperformed existing deep learning-based methods for spatial CCI inference.
• Demonstrated robustness to high noise and ability to infer both proximal and distal CCIs, as shown in a human breast cancer dataset.

Conclusions:

• SpaGraphCCI is an effective deep learning tool for inferring spatial cell-cell interactions from multimodal SRT data.
• The method successfully integrates gene expression and image features for improved CCI detection.
• SpaGraphCCI offers a practical solution for researchers studying tissue homeostasis, development, and disease progression.