Skip to main content

Advertisement

Figure 4 | BMC Systems Biology

Figure 4

From: ENNET: inferring large gene regulatory networks from expression data using gradient boosting

Figure 4

The analysis of the sampling rates s s and s f for DREAM 4 size 100 challenge: AUPR, AUROC, and a loss. The analysis of the sampling rates s s and s f for DREAM 4 size 100 challenge. A: For each set of parameters (s s , s f , M, ν)  {0.1,0.3,0.5,0.7,1} × {0.1,0.3,0.5,0.7,1} × {5000} × {0.001} we analyzed an area under the Precision-Recall curve (AUPR) in function of an average 5-fold cross-validated loss over all the observations (across all gene selection problems) from all 5 networks. For each network AUPR is decreasing in a function of a loss. For each network a point corresponding to the default set of parameters is highlighted, i.e., (s s , s f , M, ν) = (1,0.3,5000,0.001). Usually, the default set of parameters gives the minimal loss (maximal AUPR). B: By analogy, different choices of parameters lead to a different area under the ROC curve (AUROC). The two measures are consistent with each other.

Back to article page