Figure 5.

Prediction of autoimmunity with normoglycemia or T1D onset prior to seroconversion by machine learning analysis

(A) The panels show receiver operating characteristic (ROC) curves of peptide panels that predict normoglycemia (comparison I1) (n = 247) and T1D onset (comparison T1) (n = 49) at 6 months prior to the seroconversion. The numbers (n) of case-control pairs used at each time point are shown at the top of each ROC curve. Individual bootstrap curves are shown in gray with the mean curve given in blue.

(B) Heatmaps showing the selected proteins and their frequencies of being kept in the model over the 100 bootstrap iterations for the most important peptide features used to predict the model. The left two panels contain proteins that were selected in only one comparison, whereas the right panel shows proteins that were commonly selected. Proteins are named based on UniProt gene names.