AI-Driven Pipeline for Drug Repurposing: Multi-Omics, Functional Embeddings, and Structural Evaluations

Abstract

Drug repurposing is emerging as a vital strategy for accelerating therapeutic discovery while mitigating the risks and costs of de novo drug development. This study introduces a state-of-the-art computational pipeline combining multi-omics data integration, protein-protein interaction (PPI) network analysis, and advanced embedding techniques to systematically uncover drug repurposing opportunities.

Key to our methodology is the application of the network topology-based deep learning framework (NETTAG) with Functional Representation of Gene Signatures (FRoGS) embeddings, benchmarked against traditional Singular Value Decomposition (SVD) embeddings. FRoGS embeddings outperform SVD, delivering superior clustering of PPI networks and enhancing the identification of disease-associated genes (DAGs) with therapeutic potential. Using these methods, prioritized DAGs were matched against drug perturbation signatures to identify repurposing candidates, followed by ligand-based structural reranking of the top candidates using the AlzyFinder platform and a deep learning quantitative structure-activity relationship (QSAR) model.

As a case study, this pipeline was applied to circadian rhythm dysfunction in Alzheimer’s Disease (AD), identifying high-confidence insomnia-related drug candidates such as Quetiapine. Comparative analysis revealed significant advantages of FRoGS embeddings in capturing functional relationships, making this workflow a robust and generalizable tool for complex diseases.

This pipeline integrates cutting-edge AI techniques with multi-omics data, contributing to the ongoing efforts to bridge the translational gap in drug repurposing. Its disease-agnostic nature and scalability position it as a critical enabler for uncovering therapeutic opportunities across a broad spectrum of conditions.