Drug repurposing with graph neural networks

In the past 30 years, several new drugs have been released to treat many diseases and infections, and helped to improve and extend patients lives. However, nearly one-third of new drugs offer no advantage over older drugs. Moreover, through drug repurposing, existing drugs can be applied to novel biological targets. Recently, the most effective approaches to drug repurposing have been based on deep learning, with Graph Neural Networks (GNNs) attracting increasing attention. However, despite their potential, several challenges remain in application of this networks. They often operate on constrained knowledge graphs, with limited nodes / edge types and a restricted set of node features, or are narrowly focused on a specific disease such as COVID-19.

In this talk, I will introduce GNNs approaches and their applications in drug repurposing. I will describe a GNN model trained on a newly developed heterogeneous biomedical knowledge graph called NeDRex. Finally, I will discuss the incorporation of node features into the GNNs and their impact on the GNNs performance.