Biology and healthcare
Graph-structured data is ubiquitous in biology and healthcare, from molecular interactions to brain connectomes. Graph neural networks (GNNs) have emerged as powerful tools for learning on these complex relational structures.
Protein-protein interaction networks
Protein-protein interaction (PPI) networks represent physical contact between proteins in a cell. These interactions are crucial for understanding cellular processes and developing new therapeutics. GNNs can effectively model and analyze PPI networks to do the following:
- Predict new interactions: GNNs can learn patterns in known interactions to infer novel PPIs. For example, Gainza et al. (2020) (https://doi.org/10.1093/bioinformatics/btab154) developed a GNN model that predicts PPIs by learning geometric and chemical features of protein surfaces. Another instance is the Subgraph Neural Networks for Link Prediction (SEAL) model by Zhang and Chen (2018) (https://arxiv.org/pdf/1802.09691)...
