Prediction of Adverse Drug Reactions Based on Knowledge Graph Embedding and Deep Learning

Identifying potential adverse reactions of drugs can help doctors make clinical medication decisions. In view of the high-dimensional sparse features of previous studies and low prediction accuracy in constructing an independent prediction model for each adverse reaction, a prediction model of adverse reactions based on knowledge graph embedding and deep learning is developed, which can uniformly predict the adverse reactions covered by the experiment. On the one hand, knowledge graph and its embedding technology can fuse the correlation information between drugs and alleviate the deficiency of high-dimensional sparse feature matrix. On the other hand, the efficient training ability of deep learning can improve the prediction accuracy. In the study, drug characteristic data is used to construct a knowledge graph of adverse drug reactions; by analyzing the embedding effect of different embedding strategies, the best embedding strategy is selected to obtain the sample vector. Then a convolutional neural networks model is constructed to predict adverse reactions. The results show that the convolutional neural networks model has the best prediction effect under the DistMult embedding model and the 400-dimensional embedding strategy. The mean values of accuracy, F₁ score, recall and Area Under Curve were 0.887, 0.890, 0.913 and 0.957, respectively, which are better than those reported in the literature. The prediction model has good prediction accuracy and stability, which can provide an effective reference for safe medication.