DriverFinder: A Gene Length-Based Network Method to Identify Cancer Driver Genes

Figure 1

The flowchart of DriverFinder method. (a) Input datasets consist of somatic mutation, CNV, normal and tumor expression data, and gene-gene interaction network. A generalized additive model is performed on somatic mutation data to filter mutated genes which occurred at random due to long length. After that, the residual significant genes are combined with CNV to construct mutation data. The gene-gene interaction network is constructed by integrating prior gene-gene interaction network and Pearson correlated coefficient network. And the outlying matrix is constructed by analyzing interindividual variation in tumor and normal expression. (b) Given mutation data, outlying data, and gene-gene interaction network, the bipartite graph is obtained. The black nodes on the left indicate mutated genes and the blue nodes on the right represent outlying expression genes. (c) Candidate genes are obtained by greedy algorithm. The more outlying expression events the gene overlaps, the higher the gene ranks are. (d) Statistical test is performed on candidate genes to select important putative drivers by value < 0.05.