PREDICTING DRUG-DRUG INTERACTIONS WITH GRAPHIC NEURAL NETWORKS (GNNS)
International Research Journal of Modernization in Engineering Technology and Science · 2025
Although predicting new interactions is difficult due to the rarity of known drug-drug interactions (DDIs), identifying potential DDIs is essential for clinical treatments and drug development.Large datasets are necessary for traditional deep learning methods, but DDIs find it challenging to acquire them.In this paper, we present KnowDDI, a new graph neural network (GNN) based method that integrates data from large biomedical knowledge graphs to improve drug representations.In order to interpret predicted DDIs, KnowDDI creates subgraphs for every drug pair.Each connection in the subgraph denotes the significance or degree of similarity between drugs, even when there are no direct interactions.Experiments on two benchmark DDI datasets show that KnowDDI improves interpretability while achieving high prediction accuracy.Furthermore, KnowDDI works well even with knowledge graphs that are sparser, highlighting the importance of propagated drug similarities in situations with little data.KnowDDI provides an open-source tool for detecting putative DDIs and expanding its use to protein-protein and drug-target interactions by fusing deep learning with biomedical knowledge graphs.This helps with drug discovery, healthcare, and biomedicine.