Schematic summary of the detection of TPEs. The Endomesoderm GRN was derived from experimental data (step 1), from this topology we derive an ODE model (2) which is then repeatedly simulated with randomly sampled parameter sets under different perturbations (3). Exemplary, the possible effect of knock down of an activator (a-KD) and an inhibitor (i-KD) on a genes expression are shown. Qualitative topological perturbation effects, the sign of the difference between control and a-KD resp. i-KD indicated by the red and green arrows, are detected from these simulations (4) and these effects compared to experimental data (5). For the comparison, quantitative experimental data needs to be transformed to qualitative data (6). In order to rank the agreement between network topology and experimental data, the same method is applied to randomized networks (R). Alternatively, the resulting agreement can be compared to the agreement expected given that the correct topology is used, as described in section 'Inference of the Reliability of the Proposed Heuristic' and sketched in Figure 3.