Skip to main content
. 2020 May 1;9(5):1306. doi: 10.3390/jcm9051306

Figure 3.

Figure 3

Performances of the LR models in the training-validation sets (A) and the test set (B,C). After LASSO, LR models with downregulation or stimulation protocol and MCP-1 as predictor variables were built and applied to the training and the validation sets. Histograms of AUROC show, as expected, good predictive performance of this models on the training set (red bars in (A)). Median AUROC in the validation sets was similar to the one found with the training set (~ 0.90) but exhibited more dispersion (blue bars in (A)). When the response vector in the test set was permuted, the performance of the model was indistinguishable from that of the random model (median and mean AUROC equal to 0.5, blue bars in (C)) and very different for the predictive capability of the model on the original response vector (AUROC = 0.95, red bar in (C)).