A graph represents entities as nodes and relationships as edges. For molecules, atoms form nodes and bonds form edges carrying spatial-distance information; pairwise relationships encode three-dimensional orientation. section 7.3.2
In message-passing neural networks, nodes repeatedly exchange learned messages and update their hidden states. Pooling combines node states into a graph-level prediction, such as molecular energy, dipole, or gap. Information between distant nodes may require many rounds to propagate; virtual edges and master nodes are explored to speed communication. section 7.3.2
