Figure 6.

Predictive model performance. A support vector machine trained on computational biomarkers of treatment responsiveness correctly classified novel patient data using a five-fold cross validation approach with 100% accuracy (sensitivity 100%, specificity 100%; n = 9) (A). The distribution of the predictive biomarkers used for model prediction are shown in (B), along with the distributions for each individual predictive variable. The numbers alongside each datapoint show that subject’s ISS score at baseline.