Drug–drug interaction (DDI) prediction has received considerable attention from industry and academia. Most existing methods predict DDIs from drug attributes or relationships with neighbors, which does not guarantee that informative drug embeddings for prediction will be obtained. To address this limitation, we propose a multitype drug interaction prediction method based on the deep fusion of drug features and topological relationships, abbreviated DM-DDI. The proposed method adopts a deep fusion strategy to combine drug features and topologies to learn representative drug embeddings for DDI prediction. Specifically, a deep neural network model is first used on the drug feature matrix to extract feature information, while a graph convolutional network model is employed to capture structural information from the adjacency matrix. Then, we adopt delivery operations that allow the two models to exchange information between layers, as well as an attention mechanism for a weighted fusion of the two learned embeddings before the output layer. Finally, the unified drug embeddings for the downstream task are obtained. We conducted extensive experiments on real-world datasets, the experimental results demonstrated that DM-DDI achieved more accurate prediction results than state-of-the-art baselines. Furthermore, in two tasks that are more similar to real-world scenarios, DM-DDI outperformed other prediction methods for unknown drugs.

Drug-drug interaction prediction plays an important role in pharmacology and clinical applications. Most traditional methods predict drug interactions based on drug attributes or network structure. They usually have three limitations: 1) failing to integrate drug features and network structures well, resulting in less informative drug embeddings; 2) being restricted to a single view of drug interaction relationships; 3) ignoring the importance of different neighbors. To tackle these challenges, this paper proposed a multiview fusion based on dual-level attention to predict drug interactions (called MFDA). The MFDA first constructed multiple views for the drug interaction relationship, and then adopted a cross-fusion strategy to deeply fuse drug features with the drug interaction network under each view. To distinguish the importance of different neighbors and views, MFDA adopted a dual-level attention mechanism (node level and view level) to obtain the unified drug embedding for drug interaction prediction. Extensive experiments were conducted on real datasets, and the MFDA demonstrated superior performance compared to state-of-the-art baselines. In the multitask analysis of new drug reactions, MFDA obtained higher scores on multiple metrics. In addition, its prediction results corresponded to specific drug reaction events, which achieved more accurate predictions.