Figure 1

Testable nutrient predictions are generated from metabolic network data. Our prediction method operates via a four-step process. (A) A metabolic reaction network can be obtained from manual curation, computational inference, or a combination thereof. (B) The reaction network is converted into a constraint problem and solved for minimal nutrient sets. (C) These minimal nutrient sets are distilled into easier-to-handle “equivalence classes”: compounds A and B are in the same equivalence classes if for every nutrient set including A, an equivalent nutrient set exists with B substituted for A. (D) The equivalence classes are then evaluated by comparison with laboratory experiments.