Figure 4.

Machine learning survey: classifiers tuned and trained on the TCGA training set with tenfold cross-validation and tested on the unseen samples of the test set (left). Feature selection (right): comparison of each class recall on the TCGA test set for the mLRs trained on the complete gene set, on the four filter-based spaces and on the limma50 and limma50_BWE gene signatures. Normal-like class is excluded from the graph due to the trifling number of samples (only 5) in the TCGA test set.