Similar attempts to use biologically meaningful characteristics in GSA procedure have been presented in Yoon and Deisboeck (2009) and Kim et al. (2010). Yoon et al. used MPSA PD0325901 mw to identify network components controlling Erk responses to be either transient or sustained. For this purpose, two characteristic measures were introduced, the amplitude and the duration of the Erk signal, to split all parameter sets into binary classes. In Kim
et al. Sobol’s algorithm was applied to predict the parameters that control the characteristic, related to the delay time to cell death – a biologically-relevant quantity, which was not a state variable of the model. In both studies application of GSA techniques provided a valuable insight into Vismodegib the mechanism controlling input–output behaviour
of the networks, with potential to be used for identification of biomarkers for pharmaceutical drug discovery processes. The flowchart of our GSA procedure is presented in Fig. 2. Further we briefly outline key stages of the proposed GSA procedure and illustrate how each of them was implemented for our test system – ErbB2/3 network model. Step 1: Definition of the inputs to the method In our GSA implementation the inputs to the method include: S.1.1. A kinetic model of a signalling pathway, calibrated on a set of time-series data Because of our specific interest in identification of anti-cancer drug targets and the analysis of drug resistance, our version of GSA uses as an input a kinetic model of a signalling pathway, calibrated on a particular set of time-series data. Any model calibrated in this
way should contain a set of parameters, identified from a fitting procedure, to achieve the best match between experimental curves and relevant model trajectories. Suitable data represent time course profiles of phosphorylated proteins, registered after stimulation of the signalling with relevant receptor ligands. Our ErbB2/3 network model was calibrated on the set of time course profiles of pErbB3, pErk and pAkt registered after stimulation of PE04 cells with heregulin, found in the presence and absence of anti-ErbB2 inhibitor pertuzumab (see (Faratian et al., 2009b) and Fig. S6 in Additional File 1). Note that in general GSA does not require a calibrated model as an input, but here calibration is needed to confirm the validity of the model. However, full identifiability of the model is not required. S.1.2. Definition of a set of model parameters to perturb Depending on the purpose of the analysis the set can include either all system parameters or a particular sub-set. In our analysis of the ErbB2/3 network we perturbed all model parameters, including kinetic constants and total concentrations of the signalling proteins, with exception of the parameters corresponding to the concentration of external compounds, such as receptor ligands (heregulin-β, (HRG)) and inhibitors (pertuzumab (Per)), which were fixed at their values used in the experiments.