A class of deep learning models designed to process data structured as graphs, making them ideal for representing molecules and crystal structures. In materials engineering, they accelerate development by accurately predicting the physical and thermal properties of novel compounds based on their atomic structures.