Figure 5

T-test gene ranking of top classifiers. The genes from the best model, selected from 50-trial cross-validation, are plotted according to t-test gene ranking using box plots to show the distributions. As with most parsimonious machine learning methods, the sparse SVM (SSVM) classifier tends to select genes which are more univariately discriminative than either the pathway or pathway pair classifiers. Deeper mining of the features using IFR can help identify non-obvious sets of discriminative genes and elucidate discriminative pathways.