Figure 4: Machine-learning model to identify biomarker signatures.

(a) Schematic overview of the machine-learning approach used to develop multi-feature biomarker signatures. (b) The predictive importance of individual peptides to distinguish prostate cancer from normal controls. Pink bars represent the selected relevant peptides to build the predictor. Blue bar represents the predictive importance of serum PSA (SPSA). (c) ROC curves for diagnosis. The performance for the selected peptide signature (pink), SPSA alone (blue) and randomly selected peptides (grey) are compared. ROC curves are generated from 10-fold cross-validation. ROC curves generated from test set are in Supplementary Fig. 7. (d) The predictive importance of individual peptides to distinguish pathological stage pT3 from stage pT2. (e) ROC curve analyses for prognosis. Pictograms adapted from vector files by Dave, http://vector4free.com/vector/man-woman-sign-pictograms/ (CC BY 4.0).