Fig 1.
Effects of gene deletion on gene expression.
Schematic of gene expression levels for a target gene ‘V’ relative to wildtype when a gene for either a secondary (P) or a primary regulator (T) is deleted. All combinations of positive (activating) and negative (repressing) regulation (Pos. or Neg. respectively) are shown. Activating or repressing phosphorylation (Phos.) are indicated with closed or open circles and regulation by TFs (Reg.) are indicated with pointed and flat arrowheads.
Table 1.
Node set definitions.
Table 2.
Yeast perturbation data resources.
Fig 2.
Small example network with unresolvable ambiguity.
The true network (A), KPs (diamonds), TFs (circles) and target gene ‘V1’. The resulting mRNA outputs from simulated KO experiments are shown in (B-F) and represents the data used for inference. A graph representing the cumulative (total) influence through all pathways is shown in (G), and the inferred regulatory interactions are shown in (H). The node value color scale applies to (B-F) where colors show mRNA log fold-change values for each knockout. The Edge value color scale shows the “true” regulatory weights in (B-F) and the inferred values in (H). The knockout protein is indicated with a red border and strike-through. Dotted edges indicate the direct effects that were removed by the knockout.
Fig 3.
Edge inference assessment on simulated networks (A) with complete knowledge of regulatory interactions, and on yeast (B and C) where very limited examples were known. ROC curves illustrate performance for inferred edges pooled from all 25 simulated networks, each containing 100 nodes. Different types of regulation were assessed independently as indicated in the legend. Gene set ‘V’ refers to any gene. Area proportional diagrams for yeast inference are shown in (B). Gray areas represent the proportion of predicted interactions relative to the total possible for each type. Performance on the validation set (Known) are shown as colored areas representing the proportion of the known interactions that were predicted (or not predicted) for the two types of secondary regulation, d(P, T) in red and d(P, P) in blue. The counts of each interaction type are also shown next to each of the four areas in the two proportional diagrams. Fisher’s exact test p-values represent the chance of observing the prediction performance by random chance. Odds ratio for the number of shared GO Biological Process (BP) terms between interacting genes (C) based on the number of shared GO terms for inferred compared to uninferred edges. Dashed lines show the 95% confidence intervals of the odds ratios.
Fig 4.
A summary of d(P, T) edges is shown in (A). Counts of inferred edges for each combination of regulation mode for d(PK, T) or d(PP, T) to either a primary transcriptional activator or repressor. Counts statistically larger than expected are marked with asterisk. The top scoring d(P, T) edges with shared GO terms are shown in (B). The number of GO terms shared between secondary regulators and TFs are shown next to each edge. A dashed line indicates the edges was present in the evaluation set, i.e. known interactions. An asterisk on the shared GO term indicates a prediction with no known evidence. Boxes represent regulons for the TFs and are labeled with representative significant biological process GO terms or the process the TF is known to regulate. The size of the box represents the number of genes in the regulon.
Table 3.
Yeast edge data resources.
Table 4.
Gene ontology annotation resources.