Fig. 2.

Comparison of tuberculosis drug resistance predictive performance based on input feature set. Area under the ROC curve classification performance and 95% confidence intervals during repeated cross-validation for the MD-WDNN and logistic regression models trained on all features (common and derived mutations), MD-WDNN trained on just mutations (Common Mutations), and the trained on mutations and derived categories occurring in genes known to be resistant determinants for each drug (Preselected Mutations).