Fig. 6

Robustness-based comparison of different ensembles. a Robustness score distributions of feasible parameter vectors in Ensembles 1, 15, and 16. Each parameter vector’s robustness is computed by perturbing the ten most critical model parameters with nine distinct perturbation levels to simulate the ten most fragile phenotypes (Table 4). The total number of perturbations that do not lead to phenotype losses in these 900 simulations is recorded as the robustness score \(\hat R\) per feasible parameter vector. b-d Comparison of the maximum robustness \({\overline {R}}\) per phenotype-parameter pair among Ensembles 1 in b, 15 in c, and 16 in d. The relative robustness in Ensemble 15 or 16 is -1 (1) if the particular robustness value is lower (higher) than Ensemble 1