TriOmicNet: A Novel Multi-Layer Network Diffusion Approach Integrating Three Types of Scores to Identify Cancer Driver Genes.

Cancer is a complex disease driven by genetic mutations, and identifying driver genes is crucial for discovering key biomarkers and developing personalized therapies. In this study, we propose TriOmicNet, a multi-layer network diffusion method that integrates gene mutation, gene expression, and miRNA expression data. Unlike traditional methods that rely on a single data source, TriOmicNet combines three scoring mechanisms: regulatory potential, control ability, and multi-network diffusion. In addition, a node-scoring strategy is used to determine the seed nodes of the random walk algorithm. Comparative results with seven state-of-the-art methods show that TriOmicNet achieves competitive performance across several evaluation metrics and identifies potential driver genes that other methods overlook or rank lower. Ablation results indicate that the multi-network diffusion score provides the dominant predictive signal in some settings, while the other two scores provide complementary evidence rather than consistent F1 improvements. In the BRCA, LUAD, and PRAD datasets, TriOmicNet identified 121, 135, and 148 potential driver genes, respectively, that are not included in the benchmark driver gene databases. Overall, TriOmicNet provides an integrative framework for cancer driver gene prioritization and may support further investigation of cancer biomarkers and therapeutic targets.