Research on Joint Drug Prediction Based on Bipartite Networks
Bingyan Wang, Jianzhang Zhang, Xiu‐Xiu Zhan, Chuang Liu · 2024
Accurate prediction of drug interactions is crucial for improving drug safety and driving the development of combination therapy. In this study, we utilized a drug-target network as the foundation and employed network projection methods to construct a drug similarity network. We then created a series of classification features based on network topology, aiming to predict drug synergies, antagonisms, and discover potential features of synergistic drug combinations. Through feature selection analysis, we identified that features indicating whether drugs share direct connections in the similarity network exhibited distinct distributions between positive and negative samples, effectively reflecting drug interactions. When compared to baseline methods commonly used for predicting drug interactions, our approach achieved the best results with an optimal AUC of 0.914 and classification accuracy of 0.829. Furthermore, we correlated network features with actual drug-target relationships and found that synergistic drug combinations generally act on different targets. This research demonstrates that a network topology-based approach can effectively classify and predict drug synergies and antagonisms. In contrast to traditional methods relying on drug functionality, structure, and target genes for similarity, this approach offers computational simplicity and efficiency, holding the potential to drive further developments in the field of combination therapy.