Summary of predictive performance per dataset when using gene-expression and clinical predictors and performing feature selection.

We predicted patient states using gene-expression and clinical predictors (Analysis 5). Using each respective training set, we performed feature selection for each of 14 feature-selection algorithms and performed classification using n top-ranked features. For each combination of dataset, class variable, and classification algorithm, we calculated the arithmetic mean of area under the receiver operating characteristic curve (AUROC) values across 5 (outer) iterations of Monte Carlo cross-validation. Next, we calculated the minimum, first quartile (Q1), median, third quartile (Q3), and maximum for these values across the algorithms. Finally, we sorted the algorithms in descending order based on median values. Each row represents a particular dataset/class combination. (XLSX)