Link prediction in drug-target interactions network using similarity indices

Area of Science:

• Computational biology
• Bioinformatics
• Network science

Background:

• In silico drug-target interaction (DTI) prediction is crucial for drug repositioning.
• Network-based DTI prediction methods often rely on machine learning (e.g., RBM, SVM) requiring extensive drug/target characteristic data.
• Existing methods perform poorly when such characteristic data is unavailable.

Purpose of the Study:

• To propose a novel DTI prediction method utilizing only network topology information.
• To address the limitations of existing DTI prediction methods that require detailed characteristic data.

Main Methods:

• Developed a new DTI prediction approach based solely on network topology.
• Compared the proposed method against the Restricted Boltzmann Machine (RBM) approach using the MATADOR database.

Main Results:

• The proposed network topology-based method achieved higher precision for high-ranking predictions than RBM.
• This improved performance was observed specifically when no information regarding DTI types was available.

Conclusions:

• Network topology-based approaches offer a more suitable strategy for DTI prediction.
• These methods are advantageous in real-world scenarios with limited prior knowledge on drug, target, or interaction characteristics.