A graph encodes entities as nodes and relationships as edges. Formally, it consists of vertices V, edges E, and features attached to nodes, edges, or the whole graph. section 11.2
Graph representations support machine-learning workflows for drug discovery:
- They convert molecules and proteins into machine-readable structures. section 11.2
- They preserve connectivity and spatial relationships that fixed-length feature vectors may require engineers to hand-craft. section 11.8
- Graph neural networks learn context-aware node representations by repeatedly aggregating information from local neighborhoods. section 11.8
Molecular and protein graphs
Represent a molecule with atoms as nodes and bonds as edges. Represent a protein as a graph in which spatially nearby residues connect, emphasizing three-dimensional proximity over sequence order. section 11.8
























