CellMsg: graph convolutional networks for ligand-receptor-mediated cell-cell communication analysis.

Area of Science:

• Computational biology
• Genomics
• Cancer research

Background:

• Cell-cell communications (CCCs) are crucial in tumoral tissues, influencing differentiation, invasion, metastasis, and drug resistance.
• Traditional experimental methods for inferring CCCs are inefficient for large datasets.

Purpose of the Study:

• To develop a computational framework, CellMsg, for efficient inference of CCCs.
• To identify high-confidence ligand-receptor interactions (LRIs) and quantify CCC strength.

Main Methods:

• CellMsg utilizes multimodal features and graph convolutional networks to identify LRIs.
• It employs a three-point estimation method combining LRIs and single-cell RNA-seq data to measure CCC strength.
• Performance was validated using benchmark LRI datasets, cross-validation, and molecular docking.

Main Results:

• CellMsg accurately identifies high-confidence LRIs with superior prediction performance and robustness compared to existing methods.
• Identified LRIs were validated through molecular docking.
• Analysis of human melanoma tissue revealed intercellular crosstalk among seven cell types.

Conclusions:

• CellMsg provides a reliable LRI database and an effective CCC strength evaluation method for single-cell RNA-seq data.
• This computational tool facilitates the deciphering of intercellular communications in biological systems.
• CellMsg is publicly available for research use.