Figure 2. Representation of alternative models obtained based on consistency with input information provided for the glycolysis benchmark pathway.

(A) Example of three alternative models evaluated using different types of restraints based on modeling of the glycolysis pathway with a subset of pathways shown. The restraints on node patterns are shown using colored lines (blue – docking restraints, green – SEA restraints, purple – chemical transformation restraints, red – restraints with unfavorable scores). Metabolites are labeled by KEGG ID and enzymes are labeled by step in glycolysis pathway. On the left, alternate model one is consistent with docking scores, but not with all SEA scores and chemical transformations. In the middle, alternate model two is consistent with the docking scores and SEA scores, but not with chemical transformations. On the right, alternate model three is consistent with docking scores, SEA scores, and chemical transformations, thus increasing the rank of the correct enzyme-substrate pairings. (B) Alternative models shown with chemical structures. (C), Ranks of correct substrate for the corresponding enzyme at each step in the glycolysis benchmark case. 1 – glucokinase, 2 – phosphoglucose isomerase, 3 – phosphofructokinase, 4 – fructose bisphosphate aldolase, 5 – triosephosphate isomerase, 6 – glyceraldehyde 3-phosphate dehydrogenase, 7 – phosphoglycerate 8 – phosphoglycerate mutase 9 – enolase and 10 – pyruvate kinase.

Figure 2—figure supplement 1. Benchmark assessment for decoy and dummy enzymes.

(A) Two decoy enzymes were included with the four enzymes in the CMP KDO-8P biosynthesis pathway, with lengths from three to six enzymes sampled. Pathway models with the best score for different number of protein pathway members. For comparison, the correct pathway is outlined in green. (B) Scores of pathway models at different pathway lengths. The top-scoring models consisted of the four known CMP KDO-8P pathway enzymes. The best score at each pathway length is shown as a blue circle, the score of the pathway model that matches the correct pathway is shown as a green triangle, and all other scores as black dots. The cutoff for good-scoring models, which is two standard deviations below the best score, is shown as a red dashed line. (C) Enzymes in the same pathway often cluster together in dendrograms constructed based on the SEA score as a distance metric. Such clustering is illustrated for glycolysis here. (D) Assessments of the three benchmark pathways for which the candidate enzyme set is incomplete. In each pathway, one enzyme is replaced with a ‘dummy’ enzyme, for which there is a lack of input information. For serine biosynthesis, the correct pathway remained the top-scoring model. For the other cases, the inclusion of the dummy enzyme lowered the correct pathway ranking, which, nevertheless, remained within the top-scoring models.