Topological feature generation for link prediction in biological networks

Area of Science:

• Computational Biology
• Bioinformatics
• Network Science

Background:

• Graph embedding methods extract information from biological networks.
• Existing methods face high computational costs and dimensionality challenges.
• Predicting potential interactions is crucial for biological research.

Purpose of the Study:

• To address computational challenges in graph embedding for biological networks.
• To introduce the Chopper algorithm as an efficient alternative.
• To improve the speed and accuracy of link prediction in protein-protein interaction (PPI) networks.

Main Methods:

• Applied the Chopper algorithm to undirected PPI networks (nervous system, blood, heart).
• Utilized feature regularization techniques to reduce data dimensionality post-embedding.
• Compared performance against state-of-the-art graph embedding methods.

Main Results:

• The Chopper algorithm significantly reduced classifier learning time.
• The proposed method demonstrated superior performance in link prediction.
• Experiments confirmed the Chopper algorithm is faster than existing methods on PPI datasets.

Conclusions:

• The Chopper algorithm offers a computationally efficient approach to graph embedding.
• This method enhances the prediction of potential interactions in biological networks.
• The approach is effective for analyzing complex PPI networks across different biological systems.