FD-MSGL: Drug Repositioning via Frequency-Domain Multi-Source Synergistic Graph Learning

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Drug repositioning accelerates the identification of novel therapeutic indications for existing compounds.
• Current methods often rely on single data sources, limiting the integration of complex biological mechanisms.
• Integrating molecular similarity, target selectivity, and regulatory pathways is crucial for effective drug repositioning.

Purpose of the Study:

• To introduce FD-MSGL (Frequency-Domain Multi-Source Synergistic Graph Learning), a novel framework for drug repositioning.
• To address the limitations of existing methods by integrating multi-scale biological information.
• To enhance the prediction accuracy and efficiency of identifying new drug indications.

Main Methods:

• FD-MSGL integrates three biological information sources: homogeneous semantic graphs (molecular similarity), heterogeneous graphs (drug-protein-disease interactions), and pathway regulation graphs.
• Frequency-domain decomposition is employed to model both local molecular recognition and global drug family patterns.
• The framework quantifies biological synergies across molecular, target, and pathway levels, balancing structural matching with therapeutic pattern consistency.

Main Results:

• FD-MSGL successfully integrates complementary evidence from chemical similarity, target selectivity, and regulatory mechanisms.
• The framework demonstrates the ability to model local and global biological patterns simultaneously.
• Empirical evaluations on three benchmark datasets show that FD-MSGL achieves competitive performance in drug repositioning tasks.

Conclusions:

• FD-MSGL offers a powerful, multi-source approach to drug repositioning by synergistically learning from diverse biological data.
• The framework advances therapeutic development by enabling more comprehensive analysis of drug mechanisms and potential indications.
• FD-MSGL represents a significant step forward in computational drug discovery and development.