Fig. 2

Workflow of stepwise machine learning framework for identifying one-, two-, and three-controller metabolite interactions. Blue circles and arrows represent metabolites and the fluxes they might interact with, respectively, in the training set. Orange circles and arrows represent metabolites and the fluxes they might interact with, respectively, in the testing set. In each step, the training set is used to train the machine learning classifier for fluxes with a specific number of metabolite controllers, which is then applied to the testing set to predict which fluxes are in that category. Between each step of the workflow, fluxes that have been positively identified are removed from further consideration in the test set