Fig. 1.

Predictive model based on 20 compounds selected for their identification confidence and known biological role (see Fig. 4). Balanced accuracy and AUC for the specific train / test split are in the 0.70 s and 0.80 s, respectively. ROC 95% confidence intervals were calculated with 2000 stratified bootstrap replicates on the test data and are presented as blue shading around the mean curve. The balanced accuracy calculated using Monte Carlo cross-validation for these models is 0.716 for severity and 0.655 for outcome. Cross-validated AUC was calculated as ~ 0.79 in both conditions