Robustness as a measure of plausibility in models of biochemical networks

Professor Hamid Bolouri, University of Hertfordshire, Neural Systems group, Science and Technology Research Center (STRC)

The combination of Comparative Genomics with large- scale gene and protein assays promises to reveal large numbers of interspecies-conserved biochemical networks. Modeling is fundamental to understanding these incompletely-specified systems (& to engineering successful intervention strategies). In consequence, there is an increasing need for model selection and refinement strategies for biochemical networks.

We argue that robustness to reaction kinetics can be used:

* as a measure of model plausibility; * to distinguish between alternative models, and * to pinpoint potential weaknesses in a model.

2 models of the Xenopus cell cycle oscillator are used as exemplars: an old model known to have shortcomings (which our analysis highlights), and a new more accurate model (which we show can be improved using our methodology). We use a combination of bifurcation analysis and parameter optimization to study the shape of the parameter space of each model and show that sensitivities arise when either our understanding of the system is incorrect/incomplete or when a kinetic parameter is in fact a control input to the system.

Recently, von Dassow et al, and Alves & Savageau proposed statistical methods for the evaluation of robustness in biochemical models. The method presented here characterizes the parameter space of models in greater detail to permit structural comparison of alternative models . The greater quantitative detail, allows us to judge plausibility and identify structural weaknesses in a manner previously not feasible.