Figure 7. Receiver Operating Characteristic (ROC) curves
analysis for the predictive model with a combination of metabolites
calculated from the logistic regression analysis using different
metabolites selected by the VIP (> 1.0) values respectively
for a) COVID-CTRL and c) COVID-RECOV. Predictive accuracy plot of
biomarker models with an increasing number of features to discriminate
between b) COVID and CTRL and d) COVID and RECOV. The most accurate
biomarker model is highlighted by the red dots. Predicted class
probabilities based on a structure of a 2 × 2 matrix to assess the
performance of a classification model: e) COVID-CTRL groups; and f)
COVID-RECOVERED groups.
The boxplots of the most important variables (VOMS) for discriminating
between COVID and CTRL groups, and between COVID and RECOV, were plotted
(Figure 8).