GrEDeL: A Knowledge Graph Embedding Based Method for Drug Discovery From Biomedical Literatures

• Published in: IEEE access Vol. 7; pp. 8404 - 8415
• Main Authors: Sang, Shengtian, Yang, Zhihao, Liu, Xiaoxia, Wang, Lei, Lin, Hongfei, Wang, Jian, Dumontier, Michel
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Biological system modeling; biomedical knowledge graph; deep learning; Diseases; Drug discovery; Drugs; Embedding; Graphical representations; Knowledge bases (artificial intelligence); Knowledge engineering; Neural networks; recurrent neural network; Recurrent neural networks; Semantics; Unified modeling language
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2018.2886311

Drug discovery is the process by which new candidate medications are discovered. Developing a new drug is a lengthy, complex, and expensive process. Here, in this paper, we propose a biomedical knowledge graph embedding-based recurrent neural network method called GrEDeL, which discovers potential drugs for diseases by mining published biomedical literature. GrEDeL first builds a biomedical knowledge graph by exploiting the relations extracted from biomedical abstracts. Then, the graph data are converted into a low dimensional space by leveraging the knowledge graph embedding methods. After that, a recurrent neural network model is trained by the known drug therapies which are represented by graph embeddings. Finally, it uses the learned model to discover candidate drugs for diseases of interest from biomedical literature. The experimental results show that our method could not only effectively discover new drugs by mining literature, but also could provide the corresponding mechanism of actions for the candidate drugs. It could be a supplementary method for the current traditional drug discovery methods.