Algorithm 1 Integrative model based on module-network for cancer subtypes

Input: CNV data and gene expression data of two subtypes

Output: A short list of gene sets

The 1th step: Difference analysis EMDSort $(P,Q,f_{ij},d_{ij})$

(a) compute the $EMD(P,Q)$ using $f_{ij}$ and $d_{ij}$ according to the Formula (4). $f_{ij}$ is the flow and the $d_{ij}$ is the Euclidean distance.

(b) compute the $FD{R}_{ji}$ according to the emd-values.

(c) compute the q-value according to the $FD{R}_{ji}$.

The 2th step: Initial modules construction

(a) fit two normal contributions by k-means clustering and select the threshold T for each modulator.

(b) split the expression of the target gene into two sets $(A,B)$ according to the threshold T.

(c) Given a leaf vector $leaf$, the parameters $\alpha$ and $\lambda$, the size of $Leaf$ N.

(d) compute the $Score(target\_gene,modulator)$ of the split using the Formula (9).

(e) assign the target gene into the single highest scoring candidate modulator.

The 3th step: Module network learning

repeat

(a) search for a regulation program for each module.

(b) reassign each gene to the module whose program best predicts its behavior.

(c) compute the proportion of re-assigned genes $pro$.

until ($pro<0.1$)

The 4th step: The identification of candidate driver genes.