Fig. 1

Patient stratification from cross validation. The patient stratification plot shows the whole blood classifier score for each patient for (a) a representative 10-fold cross validation matching the median AUROC of 71 % across all cross validation repeats, and (b) leave-one-batch-out cross validation. Patients were sorted by classifier score and colored by their clinical response calls. The dotted lines represent the classifier threshold for predicting non-responders. No classifier threshold is shown for 10-fold cross validation because patient scores result from 10 different models (one from each cross validation fold) with 10 different score thresholds. The data set of origin is indicated for each sample at the bottom of each bar