MOTIVATION: To obtain meaningful predictions from dynamic computational models, their uncertain parameter values need to be estimated from experimental data. Due to the usually large number of parameters compared to the available measurement data, these estimation problems are often underdetermined meaning that the solution is a multidimensional space. In this case, the challenge is yet to obtain a sound system understanding despite non-identifiable parameter values, e.g. through identifying those parameters that most sensitively determine the model's behavior. RESULTS: Here, we present the so-called divide-and-conquer approach--a strategy to analyze underdetermined biochemical models. The approach draws on steady state omics measurement data and exploits a decomposition of the global estimation problem into independent subproblems. The solutions to these subproblems are joined to the complete space of global optima, which can be easily analyzed. We derive the conditions at which the decomposition occurs, outline strategies to fulfill these conditions and--using an example model--illustrate how the approach uncovers the most important parameters and suggests targeted experiments without knowing the exact parameter values.
MOTIVATION: To obtain meaningful predictions from dynamic computational models, their uncertain parameter values need to be estimated from experimental data. Due to the usually large number of parameters compared to the available measurement data, these estimation problems are often underdetermined meaning that the solution is a multidimensional space. In this case, the challenge is yet to obtain a sound system understanding despite non-identifiable parameter values, e.g. through identifying those parameters that most sensitively determine the model's behavior. RESULTS: Here, we present the so-called divide-and-conquer approach--a strategy to analyze underdetermined biochemical models. The approach draws on steady state omics measurement data and exploits a decomposition of the global estimation problem into independent subproblems. The solutions to these subproblems are joined to the complete space of global optima, which can be easily analyzed. We derive the conditions at which the decomposition occurs, outline strategies to fulfill these conditions and--using an example model--illustrate how the approach uncovers the most important parameters and suggests targeted experiments without knowing the exact parameter values.