Figure 4.

Graph neural network structure of GGraphSAGE framework. In the GGraphSAGE model, the feature matrix is aggregated around nodes through a GAT layer and two GraphSAGE layers for aggregating node, and each node contains the 3rd-order neighborhood information of the node. The final output is that each node (gene) is assigned a two-dimensional vector by the softmax layer, which consists of the probability that the node is a driver cancer gene and the probability that the node is a passenger gene. A label (1 or 0) is set to each node by comparing the magnitude of the two probabilities.