Figure 2

Simulation schemes in the CellNOptR and add-ons packages. Adapted from[40]. This network is a simplified version of a realistic toy example from[40], used to generate simulated data (triangles). We show a subset of the results of training this model to data using each of the logic formalisms available through our packages (dashed red lines). The model contains canonical pathways downstream of EGF and TNFα. The data includes: (i) a slow negative feedback from ERK to SOS-1 leading to a transient activation of ERK, (ii) a feedback from NFκ B to its inhibitor Iκ B, leading to oscillations of NFκ B, and (iii) a partial activation of p38 under combined EGF-TNFa stimulations. The CellNOptR simulation scheme (Boolean, steady state) captures the activation of ERK upon EGF stimulation (black edges EGF - SOS-1 - ERK) but not its transient nature. The Boolean with two steady states version does capture the transient ERK activation (i.e. both the black path between EGF and ERK and the negative ’AND’ gate when both EGF and ERK are activated, blue edges) but not the NFkB oscillations and p38 partial activation. With the discrete time updating scheme with Boolean state from CNORdt, we capture both the transient activation of ERK and the NFκ B oscillations(orange edges) but not the partial activation of p38. CNORfuzzy implements a continuous representation of states but with a single steady state. Thus, it captures the partial activation of p38 (pink edges) but not the behaviors of ERK and NFκ B. CNORode is based on a continuous representation of both time and state, which captures the behaviors of ERK, p38 and NFκ B (green edges). Depending on the available data and the suspected behaviors to capture, different logic formalisms are more appropriate. Dashed lines=time points used for steady states. Color of model edges: black=captured by all approaches, blue = CellNOptR(2t), orange = CNORdt, pink=CNORfuzzy, green=CNORode.