Figure 4

Receiver operator characteristic (ROC) curve for MILO generated models using 11 preselected antigens vs 23 antigens (out of 31-plex). The best model for 11 preselected antigen panel was a neural network model and the best model for 23 antigen panel (found through MILO’s unsupervised arm) was a logistic regression model as shown in the ROC-AUC graphs.