An In-Depth Analysis of Graph Neural Networks and Machine Learning Approaches for Drug Repositioning

Drug repositioning aims to identify new therapeutic applications for existing drugs offering a faster and more cost-effective alternative to traditional drug discovery. Since approved drugs already have known safety profiles, this approach is especially valuable in urgent situations like pandemic. In this study, a computational method was explored for drug repositioning using both graph-based representation for Graph Neural Networks (GNN) and feature-based representations for Machine Learning (ML) classifiers. Both models were trained separately, and their prediction scores were combined to form an integrated model named TwinNetDR. This combined approach achieved the best performance, with a precision of 95.92%, outperforming the individual GNN and ML models. The results demonstrate the benefit of combining graph and feature-based learning for reliable drug repositioning.