Fig. 4

Comparing regulatory components to their associated regulon suggests the dimension selected by OptICA more accurately models the TRN. A The average F1-scores for PRECISE 1.0, StaphPRECISE, and the B. subtilis dataset at each dimensionality. B The average F1-scores for the PRECISE 2.0 dataset at each dimensionality. C The higher dimension selected by this new method applied to the B. subtilis dataset resulted in higher mean F1-scores and a substantial increase in components with perfect scores (exact precision and recall between the component and associated regulon). D OptICA improved F1-scores and resolved under decomposition by splitting merged components computed at the PC-VA dimension. For example, the originally published decomposition reported the NadR/BirA iModulon which contained genes belonging to both the NadR and BirA regulons. The dimension selected by this new method computed separate components which more accurately model the underlying regulatory mechanisms