Drug Effect Classification Using Frequency-Based Graph Traversal Approach.

Area of Science:

• Computational biology
• Pharmacology
• Network science

Background:

• Classifying drugs as symptomatic (SYM) or disease-modifying (DM) is crucial for understanding therapeutic effects and drug repurposing.
• Existing computational methods often overlook the drug's impact on disease progression, focusing primarily on drug-target interactions.

Purpose of the Study:

• To develop a graph-based strategy for classifying drugs as SYM or DM based on their effect on disease treatment.
• To enhance drug repurposing by accurately categorizing drug mechanisms.
• To improve the interpretability of drug classification models.

Main Methods:

• Construction of a heterogeneous network integrating genes, diseases, and drugs.
• Application of a guided shortest path traversal framework to identify recurrent genes in drug-disease metapaths.
• Classification of drugs based on the presence of recurrent genes associated with specific treatment types.

Main Results:

• The proposed graph-based method significantly outperforms advanced machine learning and deep learning techniques in drug classification accuracy.
• Identification of recurrent genes provides biological insights into drug mechanisms.
• A case study on multiple sclerosis demonstrates the biological relevance and effectiveness of the approach.

Conclusions:

• The developed graph-based strategy offers a robust and interpretable method for classifying drug effects.
• This approach holds significant potential for advancing drug repurposing and understanding disease treatment mechanisms.
• The study provides publicly available data and scripts for reproducibility and further research.