MOESM9 of Development of an objective gene expression panel as an alternative to self-reported symptom scores in human influenza challenge trials

Additional file 9: Figure S6. VSURF feature selection of most predictive features. A, A random forest model was trained with only the two features ISG15 and CXCL10 as selected by VSURF. Predicted DSS is shown on the x-axis and observed DSS on the y-axis. B, Prediction of Any Symptom (DSS > 0) or No Symptom (DSS = 0) classes using the linear classifier Partial Least Square (PLS). Decision boundaries of the trained model are indicated as well as the samples used for training and testing. C, ROC curve indicating the accuracy of the PLS model.