br driver by iMaxDriverW are reported by
driver by iMaxDriverW are reported by none of the other fifteen computational tools. The iMaxDriverW reported 43, 97 and 43 new driver 1219168-18-9 in BRCA, LUSC and COAD respectively. Additionally, more than 41% of the genes classified as driver by iMaxDriverN are reported as driver by none of the other fifteen computational tools.
Furthermore, the iMaxDriverN reported 41, 68 and 37 new driver genes in BRCA, LUSC and COAD respectively. This means many new driver genes introduced by iMaxDriver. The Venn diagram of the result of gene overlaps among iMaxDriver and union of other fifteen computational tools is shown in Fig. 6. Further, in case of precision and recall iMaxDriver provided acceptable results that is available at Supplementary Fig. S1, S2.
Figure 6. The Venn diagram for correctly-predicted CDGs using iMaxDriver and the union of correctly predicted CDGs using all other methods
The current computational tools available for CDG prediction rely heavily on mutation rates. First group of methods including ActiveDriver , CoMDP , Dendrix , e-Driver , OncodriveCLUST , OncodriverFM  and Simon  exploit mutation profiles as main feature for CDG prediction. The second group of methods including
DawnRank , DriverNet , iPAC , MDPFinder , MeMo  and NetBox  rely on mutation data integrated with other omics data such as GE, network and pathways data etc. MSEA  exploits disease-related data in addition to functional genomics and gene network data and finally MutsigCV  relies on GE and exome sequence for CDG prediction. In contrast, the iMaxDriver exploits GE data together with TRN data without using mutation data. This strategy is shown to result in better F-measure values in comparison with the other state-of-the- art methods. We proposed iMaxDriverW and iMaxDriverN for weighted and non-weighted networks, respectively.
The arbitrary values (0.2, 0.5 and 0.8) is used for mapping confidence of relationships for the iMaxDriverW. Changing these values only affects the final results slightly (Supplementary Fig. S3). The random values are used for edge weighting when the network was non-weighted due to lack of information about effect of each transcription factor on the other genes. Furthermore, our results suggest that average of coverage values is a good proxy for the influence of a CDG on the regulation of other genes in iMaxDriverN. The results indicate that iMaxDriverW surpass
iMaxDriverN in number of correctly classified CDGs and F-measure since more information has used as input in
iMaxDriverW and consequently the results of IM is improved.
The CGC project aims to catalogue genes which causally result in cancer. The results of iMaxDriver method are compared with CGC driver gene list and the precision, recall and F-measure of the iMaxDriver is compared with other fifteen computational tools available for CDG prediction. The results indicate that often the recall (and sometimes the precision) of iMaxDriver are one of the best among the tested methods. However, in case of all cancer tissues, the F-measure of iMaxDriver is superior compared to all of the other methods (Fig. 4). Furthermore, iMaxDriver shows a relative “orthogonality” to othe r methods, as a considerable portion of its predicted CDGs are not detected any of the other methods. This indicates that iMaxDriver can find complementary driver genes for state-of-the-art computational CDG prediction tools.
 D. Kempe, J. Kleinberg, and É. Tardos, “Maximi zing the spread of influence through a social network,” in Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining, 2003, pp. 137–146.
The authors declare that they have no competing interests.
Data is presented within the manuscript and the Supplemental Materials.
The software is available publicly at https://github.com/majid-rh/iMaxDriver
Algorithm 1. The iMaxDriverW algorithm
Input: A directed and weighted graph of genes and a list of threshold value for each gene
ID that constructed according to descriptions provided in Subsection 3.1
Output: Result: A list of genes sorted by their coverage
1. Result ← an empty set of
2. AllNodes all nodes of G
3. Calculate MaxIncome as maximum of sum of weights of incoming edges to the nodes