Figure 4

Statistical significance of classifier performance. (a) LOOCV ROC curves for all four models based on the original dataset as well as a control obtained via a single random permutation of the R5/DM class labels among the V3 loop variants in the dataset. (b) Distribution of 10-fold CV SVM prediction performance (BAR - balanced accuracy rate and MCC - Matthew's correlation coefficient) over 1,000 permutations (random shuffles) of the class labels, compared with similar measurements obtained using the class label arrangement of the original dataset.