Figure 2. Pareto front produced by METRADE when maximizing succinate, acetate and biomass production through the parallel optimization algorithm.

The optimization algorithm starts from an array of gene expression that, when translated into flux bounds, gives the default lower and upper bound of the initial model. On low succinate, slight variations of succinate flux cause step variations of biomass (plot (f)). The initial perturbation (e,f) is applied in anaerobic conditions on the first candidate strains and improves the convergence of the algorithm, thus permitting to avoid local maxima and to increase the coverage of the objective space. As a result, we discover a new area not explored by the algorithm applied in (c,d), including a new maximum for succinate production. This technique allows detecting the subspace where the bacterium operates (also called metabolic potential) in the objective space, and investigating scenarios of adaptability over time. Solutions are denoted by progressively warmer colors according to the time step of the PGA in which they have been generated adaptively from the starting point.