Figure 4

Distributions of the E. coli training data in word space. The training data for E. coli for seven directed models is visualized in a 3-dimensional subspace of word space. The three chosen words were found to be most discriminative according to a word ranking method. Every color is associated with a different model. The point which is occupied by E. coli is also indicated. The axis correspond to words which can be associated with sets of subgraphs. If a network has a non-zero word value it must possess at least one of these subgraphs.