A. baylyi metabolic model refinement process. A. baylyi metabolic model was iteratively refined in three steps using datasets of experimental results. The initial reconstruction iAbaylyiv1 was assessed and improved using dataset 1; the resulting model iAbaylyiv2 was then assessed and refined using dataset 2, yielding iAbaylyiv3 which was again evaluated and refined using dataset 3, leading to the final model iAbaylyiv4. Since only mutants corresponding to dispensable genes in dataset 2 could be phenotyped in dataset 3, gene essentialities revealed in dataset 3 are medium-specific, i.e. conditionally essential. Genes classified as conditionally essential in dataset 3 are conditionally essential on at least one environment. Genes classified as dispensable are dispensable on all tested environments. Model accuracy figures indicates for each dataset and its corresponding models the counts of consistent and inconsistent predictions. Accuracy is computed as the fraction of consistent predictions among all predictions. For dataset 1, Biolog results for metabolites that were not in the model were counted as consistent with predictions if the metabolite was not a carbon source and inconsistent if the metabolite was a carbon source. Model corrections figures summarize the corrections performed on each model component.