VGAE-CCI: variational graph autoencoder-based construction of 3D spatial cell-cell communication network.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Cell-cell communication is vital for biological functions, development, and immune responses.
• Spatial transcriptomics sequencing (ST-seq) enables detailed analysis but faces challenges with data incompleteness and biases.
• Existing methods often overlook multi-layer and 3D tissue communication networks.

Purpose of the Study:

• To develop a novel deep learning framework, VGAE-CCI, for robust cell-cell communication inference.
• To address limitations of ST-seq data, including missing values and systematic biases.
• To enable comprehensive analysis of cell-cell communication across multiple tissue layers and in 3D.

Main Methods:

• Proposed VGAE-CCI, a deep learning framework utilizing a Variational Graph Autoencoder.
• Applied the model to spatial transcriptomics data with inherent incompleteness.
• Enabled cell clustering at single-cell resolution using spatial encoding for enhanced communication inference.

Main Results:

• VGAE-CCI demonstrated superior performance in predicting cell-cell communication across six diverse datasets.
• The model effectively handled incomplete ST-seq data, improving accuracy and reliability.
• Successfully identified cell-cell communication networks across multiple tissue layers in 3D.

Conclusions:

• VGAE-CCI offers a reliable and efficient solution for analyzing complex cell-cell communication networks.
• The framework enhances the understanding of biological processes by accurately inferring spatial cell interactions.
• This method advances the analysis of spatial transcriptomics data for biological discovery.