Novel drug discovery and development has extremely high costs and slow pace, which makes drug repurposing an attractive proposition because it involves the use of de risked compounds, with potentially lower overall development costs and shorter development timelines. Since biological experimental methods for drug repurposing also cost a lot of money and time, researchers mainly use computational methods for drug repurposing. In recent years, network-based and knowledge graph-based methods have been used widely in the area.
In this work, we are developing an end-to-end method for drug repurposing using GCNs. Our method includes three steps: knowledge graph construction, feature extraction and relationship prediction. In the first step, we use a heterogeneous knowledge graph containing information of drugs, diseases, genes, pathways, side effects, etc. We then embed nodes of the graph using a GCN. Finally, a bilinear decoder is used to reconstruct drug-disease adjacency matrix for relationship prediction.