Fig. 3.

Comparison of ProMS, ProMS_mo, and SPCA. The performance of each feature selection method was evaluated by the ability of trained logistic regression models to classify the problem specific labels (as indicated by the area under ROC curve; AUROC). A and B, performance for MSI status prediction in CRC. C and D, performance prognosis prediction in HCC. k: number of features selected. A and C, performance in the set-aside cross-validation data from the same cohort. B and D, performance in the independent data. All results are based on 100 repeats of Monte Carlo cross-validations. ns: p > 0.05; ∗p ≤ 0.05; ∗∗p ≤ 0.01; ∗∗∗p ≤ 0.001; ∗∗∗∗p ≤ 0.0001 (Wilcoxon rank sum test).