Fig. 3

Prediction of case–control status in prospectively collected blood samples using a PAM classifier. a The receiver operating characteristic curve and the corresponding area under the curve (AUC) statistics for the PAM classifier applied on the validation cohort, against a background of 100 label-shuffled control datasets that were subjected to the same model training and testing process. A t-distributed stochastic neighbour embedding (t-SNE) plot was generated using the 49 genomic regions used in the PAM classifier, coloured by b case–control status and c length of time from sample collection to diagnosis (by quartile). d Schematic of the classification results from the final PAM model on the held-out validation set alongside length of time to diagnosis (quartiles)