Embedded Methods
We trained two classifiers with the underlying algorithms SVM and GLMNET (see Table 1). Both algorithms achieved 100% accuracy (ACC) on the test set, with a 95% likelihood that the true value lies between (54% - 100%); the wide range is due to the limited size of the test set. The No Information Rate (NIR) was 0.5, as we started from a balanced dataset, and the p-value for ACC > NIR was 0.01563. We concluded that both algorithms generalised successfully, as high performance was achieved on both the training and test datasets (see also Supplemental Information).