PyMulSim: a method for computing node similarities between multilayer networks via graph isomorphism networks.

Area of Science:

• Bioinformatics
• Computational Biology
• Network Science

Background:

• Biological systems are complex, involving heterogeneous entities interacting across multiple contexts, best modeled by multilayer networks.
• Current methods for node similarity primarily focus within single networks, lacking capabilities for cross-network comparisons crucial for tasks like network alignment.
• Existing approaches struggle with evaluating nodes outside their native network structure, especially within multilayer network contexts.

Purpose of the Study:

• To introduce pyMulSim, a novel computational method for calculating pairwise similarities between nodes from distinct multilayer networks.
• To enable cross-network analysis for applications such as inferring interactions in less-studied biological networks based on well-characterized ones.

Main Methods:

• Utilizes a Graph Isomorphism Network (GIN) for representative learning of node features within multilayer networks.
• Processes learned node embeddings to compute similarity scores between pairs of nodes across different networks.

Main Results:

• Demonstrates pyMulSim's effectiveness in evaluating similarities between biological objects in source and target multilayer networks.
• Performance validated through experiments across varying noise levels and statistical significance analyses.

Conclusions:

• PyMulSim provides a reliable method for computing cross-multilayer network node similarities using GIN-based embeddings.
• The method shows high reliability and performance for comparative network analysis in bioinformatics.