Protein2Vec: Aligning Multiple PPI Networks with Representation Learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Protein-Protein Interaction (PPI) network alignment is vital for understanding biological mechanisms like protein homology and evolutionary pathways.
• Current methods often align all proteins, overlooking the biological reality that not all proteins have homologs across species.

Purpose of the Study:

• To propose a novel protein-protein interaction network alignment method that selectively maps proteins with high similarity across multiple species.
• To improve the biological relevance and accuracy of cross-species network alignment.

Main Methods:

• Integrated topological network features with biological characteristics for protein similarity assessment.
• Applied representation learning to generate low-dimensional vector embeddings for proteins, capturing surrounding structural features.
• Developed a new topological evaluation measure to assess the structural quality of alignment results.

Main Results:

• The proposed method successfully maps proteins with the highest similarity across different species' PPI networks.
• Vector embeddings effectively quantify topological similarities between proteins from distinct networks.
• New topological evaluation metric provides better insights into alignment structural quality.

Conclusions:

• The novel alignment method, integrating topological and biological features, demonstrates superior performance compared to existing methods.
• This approach offers a more biologically plausible and accurate way to perform multiple PPI network alignment.
• The findings suggest a promising direction for comparative interactomics and evolutionary studies.