GLIDE: combining local methods and diffusion state embeddings to predict missing interactions in biological networks.

Area of Science:

• Bioinformatics
• Computational Biology
• Network Science

Background:

• Biological networks, such as protein-protein interaction (PPI) networks, are crucial for understanding cellular mechanisms.
• Link prediction is a key challenge due to the inherent noise and incompleteness of experimentally derived interaction data.
• Existing link prediction methods, often developed for social networks, may not fully capture the unique structural properties of biological networks.

Purpose of the Study:

• To address the link prediction problem in biological networks by developing a novel method.
• To design a method that effectively integrates both global and local network topology information.
• To evaluate the performance of the proposed method on curated human biological and yeast PPI networks.

Main Methods:

• Developed a new embedding-based link prediction method named global and local integrated diffusion embedding (GLIDE).
• GLIDE generalizes the diffusion state distance to capture global network structure.
• Incorporated network type-specific customized measures to capture local network structure.

Main Results:

• GLIDE demonstrated effectiveness in predicting missing links in biological networks.
• The study found that local network structure dominance varies across different biological network types.
• GLIDE's global embedding measure proved valuable, particularly in regions outside highly connected network cores.

Conclusions:

• GLIDE offers an improved approach for link prediction in biological networks compared to existing methods.
• The method's ability to integrate global and local features makes it adaptable to diverse biological network structures.
• GLIDE has potential applications in identifying novel gene-disease associations, as exemplified by predictions for Crohn's disease.