DGCL: A Contrastive Learning Method for Predicting Cancer Driver Genes Based on Graph Diffusion
摘要
Gene mutations are crucial in cancer development. We introduce a method called Diffusion Graph Contrastive Learning (DGCL) to accurately identify cancer driver genes. This approach involves generating a diffusion network from protein-protein interaction networks, extracting features using graph convolution, and eliminating network noise through contrastive learning. We also constrain features through various tasks and ultimately predict driver genes using logistic regression. Experimental results demonstrate DGCL's effectiveness across various cancer types, offering a new perspective on understanding cancer mechanisms.