Optimizing Global Network Alignment With a Genetic Algorithm: Leveraging Pre-Trained Embeddings for Protein Sequences...

Area of Science:

• Bioinformatics
• Computational Biology
• Network Science

Background:

• Protein-protein interaction (PPI) networks are crucial for understanding cellular processes.
• Existing methods for aligning PPI networks face challenges in balancing accuracy and computational efficiency.
• Identifying cross-species network similarities and predicting protein complexes are key objectives.

Purpose of the Study:

• To introduce GA2Vec, a novel approach for globally aligning multiple PPI networks in a many-to-many fashion.
• To leverage advanced embedding techniques for reconstructing weighted PPI networks and incorporating functional similarity.
• To optimize network alignment using a genetic algorithm for improved accuracy and efficiency.

Main Methods:

• GA2Vec utilizes vector embeddings from ProtBERT, ESM-2, and ProtT5-XL-UniRef50 for protein sequences.
• Gene Ontology (GO) term embeddings from Anc2vec are incorporated to capture functional similarity.
• A genetic algorithm refines candidate clusters generated by community detection algorithms using a fitness function based on embedding similarities.

Main Results:

• GA2Vec achieves robust global network alignment across diverse biological networks (eukaryotic, prokaryotic, SARS-CoV, virus-host).
• Demonstrates effective alignment between SARS-CoV-2 and SARS-CoV-1 PPI networks.
• Balances key metrics including F1 score, cluster interaction quality (CIQ), internal cluster quality (ICQ), consistent clusters, and sensitivity.

Conclusions:

• GA2Vec offers an effective and adaptable solution for global PPI network alignment.
• The method successfully integrates sequence and functional information for enhanced network analysis.
• GA2Vec provides a robust framework for comparative network biology and understanding virus-host interactions.