Models in the wild-type ensemble can be clustered into 8 different groups based on their parameter values, each cluster roughly corresponding to a distinct mechanistic hypothesis.

Given information from a new experiment, we filter the wild-type ensemble for models that are predictive of the experiment outcome. Shown here are the sizes of clusters of models in the wild-type ensemble, and their corresponding sizes after filtering for each of six different perturbation experiments. The total number of models in each ensemble, the corresponding entropy score and information gain score are shown in the last three columns.