Spatial-Spectral Fusion Enables Drug Repositioning by Capturing Indirect and Long-Range Associations in Biological Networks.

Abstract

Drug repositioning accelerates clinical translation by identifying new therapeutic indications for approved drugs. However, therapeutic associations in biomolecular networks often exist indirectly, through transitive chains and long-range mechanisms, rather than as directly observed links. Shallow methods are confined to direct similarity and miss such indirect associations, whereas deep graph neural networks suffer from over-smoothing and lose discriminative power in highly connected networks. We propose a spatial-spectral collaborative framework. In the spatial domain, a wave-evolution process propagates similarity from local to global, capturing multi-hop transitive associations while preserving discriminative representations. In the spectral domain, network-specific spectral transforms model global connectivity for long-range dependencies over homogeneous similarity and heterogeneous drug-protein-disease networks, with the two views aligned by contrastive learning. On three benchmarks the method outperforms state-of-the-art baselines on most evaluation metrics; case studies on Alzheimer's and Parkinson's disease and molecular docking confirm its ability to recover non-explicit therapeutic associations. The source code and data are available at https://github.com/Juniper-cola/BIO_SSF. Supplementary data are available at Bioinformatics online.