Mathematical modeling is one of the basis of metabolic engineering, guiding genetic modifications through the study of metabolic fluxes. Stoichiometric models are an important tool to analyze metabolic networks, especially for non-model organisms or during initial analysis, since they linear models and essentially require the stoichiometry matrix and information on reversibility of the reactions as input. They can be used to explore different assumptions and scenarios, and elucidate some properties of the metabolism. Therefore some stoichiometric modeling techniques were implemented in a single stand-alone software and used to study the core metabolism of the bacteria Burkholderia sacchari for polyhydroxyalkanoate production, showing that they are a valuable tool for exploring how metabolisms work and guiding future experiment design. However, cellular metabolism is actually subjected to nonlinear dynamics and, therefore, nonlinear models are better suited to represent more diverse physiological states, which can result in better predictions. Mechanistic models are a class of such models; however, in the metabolic engineering context, all frameworks that have been proposed to estimate the kinetic parameters involved are prone to identifiability issues. Based on this obstacle, an investigation on regularization methods for ill-conditioned parameter estimation problems was conducted. Regularization methods based on the eigenvalue decomposition of the (reduced) Hessian matrix were shown to be optimal for linear parameter estimation, in the sense of reducing parameter variance, and helpful in dealing with nonlinear problems with nearly flat neighborhood around the solution. Moreover, the eigenvector-based regularization in both cases was able to recognize groups of correlated parameters, which allows for better understanding the underlying identifiability issues.