Fig 1.
Coding cancer drivers and genes with mutations.
Genes with driver mutations are cancer drivers. Some genes which do not bear mutations but regulate driver mutations to progress cancer are also considered as cancer drivers.
Fig 2.
(1) Building the network for a condition: (a) Prepare matched miRNA and TF/mRNA expression data, (b) Build miRNA-TF-mRNA network where nodes represent miRNAs/TFs/mRNAs and an edge between two nodes indicates there is a significant Pearson correlation between the expression of the two nodes, (c) Create the network by combining the miRNA-TF-mRNA network with the PPI network and other existing databases, and (2) Identifying coding and miRNA drivers: (a) Detect critical nodes, (b) Identify candidate cancer drivers.
Fig 3.
Determining the directions of edges in the miRNA-TF-mRNA regulatory network.
In a miRNA-TF-mRNA regulatory network, miRNAs can regulate TFs and mRNAs, TFs can regulate miRNAs and mRNAs, TFs/mRNAs can regulate other TFs/mRNAs. This motif is adapted from the work of [66]. In addition, the databases used to filter out edges of the network are shown on arrows.
Fig 4.
Characterising the controllability of the miRNA-TF-mRNA network.
(A) Identification of critical, ordinary, and redundant nodes in the network. (B) Average in-degree and accumulative in-degree distribution (i.e. the in-degree i with the probability p means that the probability to pick a node which has in-degree larger than or equal to i is p) for three different node types. (C) Average out-degree and accumulative out-degree distribution for three different node types.
Fig 5.
The cancer drivers predicted by each method are validated against CGC. Each bar in the chart indicates the number of validated coding driver genes for each method.
Fig 6.
Comparison of Precision, Recall, and F1Score for the top ranking genes predicted by OncodriveCLUST, ActiveDriver, OncodriveFM, DriverNet, DawnRank, NetSig, and CBNA.
In each diagram, the x-axis is the number of the top ranking genes. The y-axis is the value of Precision, Recall, or F1 Score.
Fig 7.
Evaluation based on the total number of predicted driver genes.
(A) Number of predicted drivers, (B) Fraction of validated drivers in the CGC and raw count of predicted drivers indicated on top of each bar.
Fig 8.
CBNA using different adjusted p-value cutoffs.
The cancer drivers predicted by CBNA with different adjusted p-value cutoffs are validated by the CGC. Each bar in the figure shows the number of validated coding cancer drivers of CBNA with a cutoff.
Fig 9.
Overlap between different methods.
The diagram shows the overlap among the four methods in their top 50, 100, 150, and 200 predicted drivers. For each of the four cases, the horizontal bars at the bottom left show the numbers of predicted cancer drivers validated by the CGC for the four methods; the vertical bars and the dotted lines together indicate the numbers of validated cancer drivers which overlap with each other.
Fig 10.
Validation using a well-curated set of breast cancer drivers.
The cancer drivers predicted by the methods are validated by a well-curated set of breast cancer drivers. Each bar in the figure shows the number of validated coding cancer drivers of each method.
Table 1.
The top 20 mutated coding drivers using mutation density.
Fig 11.
Identification of coding and miRNA cancer drivers.
The chart shows the percentage of different types of cancer drivers identified by CBNA from the BRCA dataset.
Table 2.
miRNA BRCA drivers predicted by CBNA.
Table 3.
Predicted drivers which are specific to each breast cancer subtype.
Table 4.
Top 20 coding and 17 miRNA drivers predicted for EMT in breast cancer.