Graph Neural Networks for Cancer Driver Gene Prediction: From Fundamentals to Applications.

Identifying cancer driver genes (CDGs) remains a central challenge in cancer genomics and is fundamental to precision oncology. Unlike passenger mutations, CDGs play a causal role in tumor initiation and progression. In recent years, Graph Neural Networks (GNNs) have emerged as a powerful framework for this task, as they naturally integrate heterogeneous multi-omics data within the context of protein-protein interaction (PPI) networks. This chapter provides a comprehensive technical protocol for predicting CDGs using a Graph Convolutional Network (GCN). We describe a step-by-step pipeline encompassing data acquisition and preprocessing, including the use of the STRING PPI network, TCGA-derived multi-omics data, and curated gene annotation resources. The protocol details critical stages such as feature engineering, where molecular profiles are combined with network centrality measures, and addresses the severe class imbalance inherent to CDG prediction through the use of the Focal Loss function. Finally, the chapter guides the reader through hyperparameter optimization, model training, and performance evaluation, while discussing practical considerations related to generalization, computational cost, and interpretability. Overall, this chapter serves as a practical manual for researchers seeking to apply GCN-based methods to cancer driver gene discovery.