(a)
(b)
(c)
(d)
Figure 4: MARS analysis of differentially abundant protein spots in HF subjects. Inputs to the model were protein spots that were differentially abundant at with B-H correction in HF (31 spots, ) subjects with respect to NH controls (). We employed 10-fold cross-validation ((a) and (c)) and 80% testing/20% training ((c) and (d)) approaches to assess the fit of the model for testing dataset. Shown are the protein spots identified with high ranking (score >20) by CV (a) and 80/20 (b) approaches for creating the MARS model for classifying HF from NH subjects. Protein spots in panels (a) and (b) are identified as spot number-protein name, and fold changes (increase ↑, red; decrease ↓, blue) are plotted. The ROC curves show the prediction success of the CV (c) and 80/20 (d) models. Blue curves: training data (AUC/ROC: 1.00), red curve: testing data (AUC/ROC: 0.97 for CV and 0.857 for 80/20).