Extracting drug-drug interactions with hybrid bidirectional gated recurrent unit and graph convolutional network

Drug-drug interactions are critical in studying drug side effects. Thus, quickly and accurately identifying the relationship between drugs is necessary. Current methods for biomedical relation extraction include only the sequential information of sentences, while syntactic graph representations have not been explored in DDI extraction. We herein present a novel hybrid model to extract a biomedical relation that combines a bidirectional gated recurrent unit (Bi-GRU) and a graph convolutional network (GCN). Bi-GRU and GCN are used to automatically learn the features of sequential representation and syntactic graph representation, respectively. The experimental results show that the advantages of Bi-GRU and GCN in DDI relation extraction are complementary, and that the utilization of Bi-GRU and GCN further improves the model performance. We evaluated our model on the DDI extraction-2013 shared task and discovered that our method achieved reasonable performance.