Decoding functional cell-cell communication events by multi-view graph learning on spatial transcriptomics.

Area of Science:

• Computational Biology
• Systems Biology
• Genomics

Background:

• Cell-cell communication is crucial for biological processes, mediated by ligand-receptor (LR) pairs.
• Identifying specific functional communication events (FCEs) that drive target responses remains challenging due to complex interactions.
• Existing methods struggle to disentangle factors influencing FCEs and their downstream gene expression effects.

Purpose of the Study:

• To develop a novel computational method for decoding FCEs and their target gene relations.
• To leverage spatial transcriptomic data and deep learning for understanding cell-cell communication.
• To identify specific FCEs and their impact on cellular phenotypes in complex microenvironments.

Main Methods:

• Developed HoloNet, a deep learning model integrating LR pairs, cell-type spatial distribution, and gene expression.
• Modeled cell-cell communication events (CEs) as a multi-view network.
• Employed an attention-based graph learning approach to predict gene expression and interpret FCEs.

Main Results:

• HoloNet successfully decoded FCEs in breast and liver cancer spatial transcriptomic datasets.
• The method effectively disentangled the contributions of LR signals and cell types to biological processes.
• HoloNet demonstrated superior performance in ligand-receptor prioritization compared to existing methods via simulation experiments.

Conclusions:

• HoloNet provides a powerful tool for dissecting cell-cell communication landscapes.
• The method accurately identifies vital FCEs that shape cellular phenotypes.
• HoloNet offers a robust approach for understanding gene regulation in specific microenvironments.