Figure 3.

An outline of the utilization of machine learning methods in metabolism modeling application for optimization of parameters in the model, as well as testing different input conditions. In this example, parameters in the model are selected at random (e.g., using Monte Carlo sampling) or, alternatively, reactions are taken out or put in (in the constraint-based models) and the models are run. The success of each model is determined using the parameter of interest (e.g., cell growth or cell death, production of the molecule of interest, etc.). Model parameters (1–N) are used as feature vectors with the success label used as the class label in the machine learning classifier. The classifier determines patterns in the parameter space with the highest discriminatory power that ensures success according to the model.