Fig. 1

The workflow of Dytidriver. We divided our whole process of cancer driver gene identification into four steps and marked with ‘a’,’ b’, ‘c’, ‘d’. In the step ‘a’, we filtered the mutated genes for each patient according to whether or not it influenced the expression of downstream genes. Only the mutated genes which connect at least one outlying genes would be included in our study. Then, the filtered mutated genes for all patients were mapped to the human functional interaction network to construct the Mut-Mut matrix. The ‘b’ step is to generate the tissue-specific PCC matrix. For each cancer, we chose the top one or two tissues with the higher association score in disease-tissue matrix as the cancer related tissues such as the tissue 1 and tissue 2 for disease D1. For each tissue, we calculated its gene-gene pearson correlation values across the whole patients and then generated the gene-gene PCC matrix by keeping the absolute PCC values more than 0.3 while left setting to 0. If there are more than one tissue related to a cancer, the final tissue-specific PCC matrix is constructed by averaging the values in the gene-gene PCC matrix of each tissue. In the ‘c’ step, we constructed the ECC mutated matrix by utilizing the ECC equation. In the final ‘d’ step, we assigned each mutated gene in the network a score by summing up all the ECC values of its connecting edges and then multiply to its corresponding variation frequency. According to the scores, the mutated genes were ranked in a descending order and those ranked at the top list the were considered as potential driver genes