This thesis is devoted to the study and application of constraint-based metabolic models. The objective was to find simple ways to handle the difficulties that arise in practice due to uncertainty (knowledge is incomplete, there is a lack of measurable variables, and those available are imprecise). With this purpose, tools have been developed to model, analyse, estimate and predict the metabolic behaviour of cells.
The document is structured in three parts. First, related literature is revised and summarised. This results in a unified perspective of several methodologies that use constraint-based representations of the cell metabolism. Three outstanding methods are discussed in detail, network-based pathways analysis (NPA), metabolic flux analysis (MFA), and flux balance analysis (FBA). Four types of metabolic pathways are also compared to clarify the subtle differences among them.
The second part is devoted to interval methods for constraint-based models. The first contribution is an interval approach to traditional MFA, particularly useful to estimate the metabolic fluxes under data scarcity (FS-MFA). These estimates provide insight on the internal state of cells, which determines the behaviour they exhibit at given conditions. The second contribution is a procedure for monitoring the metabolic fluxes during a cultivation process that uses FS-MFA to handle uncertainty.
The third part of the document addresses the use of possibility theory. The main contribution is a possibilistic framework to (a) evaluate model and measurements consistency, and (b) perform flux estimations (Poss-MFA). It combines flexibility on the assumptions and computational efficiency. Poss-MFA is also applied to monitoring fluxes and metabolite concentrations during a cultivation, information of great use for fault-detection and control of industrial processes. Afterwards, the FBA problem is addressed. A possibilistic approach is derived to get predictions under the assumption that cells have evolved to be optimal (Poss-FBA). It captures alternate optima and grades sub-optimality, thus relaxing the original assumption. The last contribution is a procedure to validate constraint-based models when data are scarce. This procedure mitigates validation problems with small metabolic networks.
This thesis highlights the importance of accounting for uncertainty when modelling living cells and promotes a constraint-based perspective: if we cannot exactly model how cells operate, use the knowledge available to distinguish what is possible from what is not. Following this idea, methods are proposed that start by representing the available knowledge and its uncertainty, and then exploit this representation to generate reliable new information.