Fig. 8

PKN quality and training set size. Fitness function F = (f _T , − N _ess. nodes , N _nodes , − N _edges ) and s _all score for model networks (blue boxplots), random sub-networks of the PKN (grey boxplots) and PKN (black line) as a function of noise in the input PKN (left panel), errors in the training set (centre panel) and size of the training set (right panel). Left panel: results based on the training set 1 given in Additional file 4 and PKNs 1 to 6 (Additional file 3). Centre panel: results based on training set 1 with 0 to 50 % error (Additional file 2) and PKN 1 (Additional file 4). Right panel: results based on training sets 1, 2 and 3 (Additional file 2) and PKN 1 (Additional file 4). Each blue boxplot summarizes measurements on 50 model networks obtained out of 500 independent runs of the optimization method (more details in main text). Each grey boxplot summarizes measurements on 350 random sub-networks of the PKN used as input for the optimization method. Component N _ess. nodes of the fitness function was always optimal (N _ess. nodes  = 14) and is not shown in this figure