Comparative analysis of network-based approaches and machine learning algorithms for predicting drug-target...

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug repositioning relies on efficient prediction of drug-target interactions (DTIs).
• Various computational methods exist for DTI prediction, including network-based, machine learning, and integrated approaches.
• Assessing the comparative accuracy of these DTI prediction methods is crucial for advancing drug discovery.

Purpose of the Study:

• To provide a comprehensive overview of network-based, machine learning, and integrated DTI prediction methods.
• To evaluate the prediction performance of state-of-the-art DTI prediction techniques.
• To identify the most advantageous methods for increasing DTI prediction accuracy.

Main Methods:

• Network-based methods utilize graph-theoretic algorithms on DTI networks and similarity matrices.
• Machine learning methods employ known DTIs and feature data for predictive model training.
• Integrated methods combine network-based and machine learning techniques for enhanced DTI prediction.

Main Results:

• Integrated DTI prediction methods generally outperform network-based and machine learning approaches.
• Some existing methods exhibit low accuracy for predicting interactions of novel drugs not in training datasets.
• Combining similarity matrices via data fusion did not consistently improve prediction accuracy.

Conclusions:

• Integrated computational methods represent the most effective strategy for accurate drug-target interaction prediction.
• Further research is needed to address limitations in predicting interactions for unknown drugs.
• Future directions should focus on improving the generalizability and accuracy of DTI prediction models.