Figure 4
From: Network modeling of patients' biomolecular profiles for clinical phenotype/outcome prediction
Comparison of the average AUROC values obtained by P-Net with the Park’s, Random Forest, Naive Bayes, and SVM methods over three datasets. AUROC results are averaged across 15 rounds of the 10-fold cross-validation. Top: average AUROC with error bars indicating standard deviation (for each dataset, bars are sorted according to decreasing performance). Bottom: statistical comparison of P-Net with the other methods. P-values are computed by one tail Wilcoxon rank sum test across 15 repetitions of 10-fold cross-validation. Rows refer to the methods compared with P-Net; columns to Breast, Colorectal and Colon cancer data.
