Abstract
Cancer progression is orchestrated by the accrual of mutations in driver genes, which endow malignant cells with a selective proliferative advantage. Identifying cancer driver genes is crucial for elucidating the molecular mechanisms of cancer, advancing targeted therapies, and uncovering novel biomarkers. Based on integrated analysis of Multi-Omics data and Network models, we present MONet, a novel cancer driver gene identification algorithm. Our method utilizes two graph neural network algorithms on protein-protein interaction (PPI) networks to extract feature vector representations for each gene. These feature vectors are subsequently concatenated and fed into a multi-layer perceptron model (MLP) to perform semi-supervised identification of cancer driver genes. For each mutated gene, MONet assigns the probability of being potential driver, with genes identified in at least two PPI networks selected as candidate driver genes. When applied to pan-cancer datasets, MONet demonstrated robustness across various PPI networks, outperforming baseline models in terms of both the area under the receiver operating characteristic curve and the area under the precision-recall curve. Notably, MONet identified 37 novel driver genes that were missed by other methods, including 29 genes such as APOBEC2, GDNF, and PRELP, which are corroborated by existing literature, underscoring their critical roles in cancer development and progression. Through the MONet framework, we successfully identified known and novel candidate cancer driver genes, providing biologically meaningful insights into cancer mechanisms.