Algorithm 1.

Gene ranking of Wrapper Embedded Recursive Feature Elimination (WERFE)

Input: Input data X : x1, x1...xm and labels Y : y1, y1...ym, where m is the number of samples. x is n-dimensional gene vector.s is the step size of RFE.

Output: Ranked genes GR of all the genes.

1: for k = 1:10 do

2:    The data set was randomly divided into ten equal parts;

3:    Keep one part as a test data; The remaining nine parts are used as training data;

4:    while X is not empty do

5:       Train a model based on training data of X using SVM;

6:       Calculate the prediction accuracy of the model using the test data;

7:       Obtain the weight of each gene produced from SVM;

8:       Remove s least weighted genes and update X;

9:    end while

10:    Obtain the gene subset G1 with the highest prediction accuracy;

11:    while X is not empty do

12:       Train a model based on training data of X using RF;

13:       Calculate the prediction accuracy of the model using the test data;

14:       Obtain the importance of each gene produced from RF;

15:       Remove s least weighted genes and update X;

16:    end while

17:    Obtain the gene subset G2 with the highest prediction accuracy;

18:    Count the votes for all the genes contained in both G1 and G2;

19: end for

20: Rank genes based on votes and obtain GR.