Depending on the purpose of the analysis, various approaches have been suggested to incorporate GSA in the general pipeline of network model development and validation (Kim et al., 2010, Rodriguez-Fernandez and Banga, 2010 and Zi et al., 2008). In this study we sought to develop a GSA procedure which would be applicable to identification of the critical nodes that exhibit the most control over the output signals from cancer-related signalling networks, and therefore could be considered as candidates

for targeting with anti-cancer drugs, or as biological markers of cancer and drug resistance. Below we briefly outline the most click here popular GSA approaches currently in use, justify the choice of the techniques for our GSA procedure, http://www.selleckchem.com/products/BI6727-Volasertib.html describe the proposed algorithm and then highlight its applied aspects. In general, all global SA techniques are designed to allow exploration of the model behaviour in the space of the model input factors. Therefore, at the first stage, they employ various sampling

algorithms for extraction of parameter sets from predefined areas of parameter space. Then for each parameter set the model outputs are calculated, and various SA methods are applied to deduce particular metrics to quantitatively describe model input–output relationships. Thus, one way of classifying the existing GSA implementations would be to characterise them with regard to their choice of (1) the sampling method, (2) the method for sensitivity analysis, (3) the characteristic used to assess the parametric sensitivity. Classical “grid” approaches which would

allow one to systematically cover the parameter space with “n” points on each individual parameter direction, cannot be used in a high-dimensional space, because of the exponential increase in volume associated with adding extra dimensions to a mathematical space that results in a computationally intractable task. That is why special sampling algorithms should be employed to effectively extract the points from a high-dimensional parameter space. The most commonly used sampling methods are pure Monte-Carlo (MC), when points are taken randomly from multi-dimensional distribution (Balsa-Canto unless et al., 2010 and Yoon and Deisboeck, 2009) and Latin Hypercube Sampling (LHS) (Jia e al., 2007 and Marino et al., 2008). LHS, a variant of stratified sampling without replacement, ensures better estimation of the mean and the population distribution function compared to pure random MC sampling (Saltelli, 2004). In our GSA implementation, we used Sobol’s low-discrepancy sequence (LDS) as our sampling method (Sobol, 1998). Sobol’s LDS belongs to the class of quasi-random sampling methods, designed to systematically fill the gaps in the parameter space, rather than to select points purely randomly.