CBS Computational Biology Services

CBS has a large experience in large-scale integrative modelling. With the advent of large sets of experimental data, integrative models with a large number of reactions are becoming more popular, and currently there are in progress several projects to model the entire cell. The critical points when building large models are (1) robustness of the model, and (2) significance of the parameters. A model is called robust when its behaviour does not change dramatically upon small changes in the parameters. Large models, with inter-related control loops, are often not robust, and consequently they are not good descriptions of the real world. Parameter significance refers to how significant, i.e., how close to the true value they are. In spite of the wealth of experimental data, eventually one has to deal with parameters whose value is unknown. The estimations of these parameters is subjective and is almost an art. Nevertheless, a rule of thumb is that parameter estimations should always be done ab initio before comparing the results of the model with experiments. Fitting the model to experimental results in order to estimate an unknown parameter should only by done with the outmost care because the significance of parameters found by these fittings is at best dubious.

An example of CBS experience in large-scale modelling can be found in Free Radic. Biol. Med., 21, 917-943. The model simulates mitochondrial free radical metabolism, is composed of 776 reactions, is robust, and has a high degree of significance. At the reference steady-state, all logarithmic sensitivites have a value in module lower than 2, revealing that most perturbations are either attenuated or modestly amplified, as could be expected for a physiological response. When iron concentrations are increased to levels that are suppose to kill the cell, logarithmic sensitivites become very large, revealing a system that is not robust, as would be expected from a model that simulates the loss of cellular homeostasis. Among other results, the model suggested for the first time that perhydroxyl has a key role as initiator of mitochondrial lipid peroxidation (a prediction not yet confirmed experimentally) and that the consecutive action of PLA₂ and glutathione peroxidase is not a biological relevant pathway for the detoxification of membrane lipid hydroperoxides (a prediction later confirmed by experiments).