The rapid pace at which genome scale reconstructions are becoming available has necessitated the need to validate in silico growth predictions with in vivo data. One way of carrying out this validation is by contrasting in silico predicted growth or lackthereof with in vivo experimental data for different gene knockouts and/or different carbon sources. Specifically, by adopting an FBA based description the biomass formation is maximized subject to stoichiometry, uptake scenario and possible gene knock-outs. If the maximum flux into biomass is found to be zero then it is deemed that the in silico prediction is for no-growth (i.e., NG). Alternative if the max flux into biomass is non-zero then the corresponding in silico prediction is for growth (i.e., G). These in silico prediction are then contrasted against the in vivo experimental prediction (i.e., G or NG). This yields four possible cases: (i) GG when both model and experiment predict growth; (i) GNG when the model predicts growth but the in vivo prediction is for no-growth, (iii) NGG when the model predicts for no growth but there in vivo growth, and finally (iv) NGNG when both model and experiment predict no growth. The idea is to try to automatically reconcile GNG predictions (by removing reactions from the model) and NGG predictions (by adding reactions to the model). A bilevel optimization formulation and solution procedure is put forth for minimally resolving all observed inconsistencies. The approach is demonstrated using the E. coli model.