FIGURE 1.

The flowchart of the proposed method. From the somatic mutation matrix, the mutation frequency of each gene is obtained to select the highly mutated genes, which will be integrated by mapping into the gene expression matrix to get the integrative network. After FDR, FC, and mutual information ranking, the feature genes serve as the input to GBDT. Then, the optimal model is obtained by Bayesian optimization. During the training process, the optimal gene subset is obtained simultaneously.