Fig. 1

The dimensionality tree of PRECISE 1.0 reveals how its ICA decomposition evolves across a range of dimensions. Each point represents a computed component, and a row of points represents all components calculated at a particular dimensionality. Connections between components of adjacent dimensions were established where their correlation was greater than 0.3. A The red box highlights a section of the dimensionality tree where component splitting occurs. At low dimensions, components undergo reorganization whereby disparate gene sets, initially contained in a single component (gray star node), split into multiple components which more accurately reflect underlying transcriptional regulatory mechanisms (purple and orange star nodes). The “Gene Start” (position of gene along the genome) vs “Gene Weight” (independent component gene weight) graphs illustrate a specific example of component splitting. The gene sets contained in the XylR and CsqR iModulons are contained in the same independent component at low dimensionality but are split into separate components at higher dimensions. These split components more accurately map to a single regulon. B The red box highlights a point of the dimensionality tree at which a single gene component appears. At high dimensions new components appear which are uncorrelated to those of the preceding dimension (green star node). Often, these components contain a single highly weighted gene and signify the commencement of over-decomposition