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

Drug repositioning accelerates therapeutic development by revealing novel indications for approved compounds, yet existing methods typically rely on single information sources and fail to integrate multi-scale biological mechanisms spanning molecular similarity, target selectivity, and multi-step regulatory pathways. Here we introduce FD-MSGL, a Frequency-Domain Multi-Source Synergistic Graph Learning framework that addresses this challenge by integrating three complementary biological information sources. FD-MSGL constructs homogeneous semantic graphs to capture molecular similarity in chemical space, heterogeneous graphs to model direct drug-protein-disease interactions revealing target selectivity, and pathway regulation graphs to trace indirect therapeutic mechanisms through multi-step regulatory cascades. Through frequency-domain decomposition, FD-MSGL simultaneously models local specificity of molecular recognition and global organizational patterns of drug families. The framework quantifies biological synergies across molecular, target, and pathway levels, integrates complementary evidence from chemical similarity, target selectivity, and regulatory mechanisms, and dynamically balances precise structural matching with global therapeutic pattern consistency. Empirical evaluation across three benchmark datasets demonstrates that FD-MSGL achieves competitive performance.