The signed two-space proximity model for learning representations in protein-protein interaction networks

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Predicting protein-protein interactions (PPIs) is vital for understanding biological processes and diseases.
• Experimental methods for PPI determination are costly; machine learning offers an alternative.
• Signed PPI networks, incorporating activating and inhibitory interactions, require specialized analysis.

Purpose of the Study:

• To introduce the Signed Two-Space Proximity Model (S2-SPM) for analyzing signed PPI networks.
• To develop a model that explicitly differentiates between positive and negative protein interactions.
• To identify archetypes representing extreme protein profiles within these networks.

Main Methods:

• Leveraging two independent latent spaces to model positive and negative interactions separately.
• Representing protein similarity through proximity within these latent spaces.
• Utilizing link prediction tasks and Gene Ontology (GO) enrichment analysis for validation.

Main Results:

• S2-SPM demonstrates superior performance in predicting interaction presence and sign compared to baseline methods.
• Enrichment analysis confirms the biological relevance of identified archetypes and their associated biological tasks.
• Statistical significance, sensitivity analysis, and BNMI metric confirm model robustness and reliability.

Conclusions:

• S2-SPM accurately models complex regulatory mechanisms in biological systems by incorporating signed PPIs.
• The model provides insights into functional roles of different interaction types and identifies key protein archetypes.
• The S2-SPM is freely available, facilitating further research in signed network analysis.