Fig. 1.

(A) Flowchart of the learning process to create a classifier of wt vs mut TP53 samples. (B) Detailed description of the central learning module: For each one of the 100 iterations, we ran the learning process for every combination of C. GL (gene‐probe list) , GR (gene‐probe ranking), and ML (machine‐learning method). Each iteration uses a different learning set and corresponding ranked GL, generating a subclassifier of k = 1,2,…N ≤ 300 genes. Optimal GL*, GR*, ML*, and k* are selected on the basis of the validation error and robustness. The final model consists of the majority vote of C (70 ≤ C ≤ 100) subclassifiers based on k* genes and the selected GL*, GR*, and ML* methods. See text for detailed description.