Kinetic models are essential for studying complex behavior and properties of metabolism. However, a persistent hurdle for constructing these models is uncertainty in the kinetic properties of enzymes. Currently available methods for building kinetic models cope in an indirect way with uncertainties by integrating data from different biological levels and origins into models. In this context, we have recently proposed iSCHRUNK1 a new approach for characterization and reduction of uncertainty that makes use of the ORACLE2 (Optimization and Risk Analysis of Complex Living Entities) framework and machine learning classification techniques. ORACLE allows us to exploit synergies between different data sources and generate population of nonlinear kinetic models that are consistent with a reference physiological states and the available data. On the other hand, with the machine learning techniques we are able to data mine the integrated datasets together with the outputs of ORACLE in order to discern complex interactions and correlations between the parameters of kinetic models and available partial data such as metabolite concentrations, metabolic fluxes, thermodynamic data and information about kinetic properties of enzymes. The proposed approach is versatile and can be used to infer more accurate ranges of kinetic parameters of all enzymes in large-scale and genome-scale metabolic networks that are consistent with the observed properties of the metabolic network. In this work, we studied which enzymes determine the sign (positive or negative values) of important control coefficients, and then we identified what should be the configuration of the kinetic parameters of identified enzymes that will determine its role as a rate-limiting enzyme. Based on this analysis we categorized the enzymes in the network in groups that act on studied properties: (i) in a concerted fashion; (ii) antagonistically; or (iii) in a mutually independent way. This work will allow us to investigate how to engineer the enzymes in order to achieve a pre-specified metabolic behavior and address the inverse metabolic engineering problems in a new way.