Prediction of Cancer Driver Genes Based on Pyramidal Dynamic Mapping Algorithm.

Pi-Jing Wei,Shu-Li Zhou,Rui-Fen Cao,Yansen Su,Chun-Hou Zheng

DOI: https://doi.org/10.1007/978-981-99-4749-2_7

2023-01-01

Abstract:Cancer driver genes are involved in the development and progression of cancer. Hence, identifying cancer driver genes is vital for finding biomarkers and learning the knowledge regarding the molecular mechanisms of cancer. In order to improve the prediction accuracy of the model for driver genes, a new network framework, which is called multi-information enhancement network, is designed by combining the MLP network and Bottleneck module with a dynamic mapping structure. It introduces dynamic mapping module into three characteristic attributes of multi-omics data to better integrate multi-omics data and extract more information, so as to accurately predict cancer driver genes. First, we use three branches to extract features with different properties, including linear features, nonlinear features and high-dimensional abstract features. Then, the three features are mapped dynamically, and each mapping module recursively iterates twice. Finally, the new enhanced features are entered into the classifier. The output of the model represents the probability that a gene is a driver gene. We applied our method and four other existing methods to predict cancer drivers. The experimental results show that the area under the receiving operation characteristic curve and the recall rate curve of the proposed method are 0.9333 and 0.8452, respectively, which are significantly superior to other methods, proving the effectiveness of the proposed model.