Abstract
The collection all TFs, target genes and their interactions in an organism form a gene regulatory network (GRN), which underly complex patterns of transcription even in unicellular species. However, identifying which interactions regulate expression in a specific temporal context remains a challenging task. With multiple experimental and computational approaches to characterize GRNs, we predicted general and phase-specific cell-cycle expression in Saccharomyces cerevisiae using four regulatory data sets: chromatin immunoprecipitation (ChIP), TF deletion data (Deletion), protein binding microarrays (PBMs), and position weight matrices (PWMs). Our results indicate that the source of regulatory interaction information significantly impacts our ability to predict cell-cycle expression where the best model was constructed by combining selected TF features from ChIP and Deletion data as well as TF-TF interaction features in the form of feed-forward loops. The TFs that were the best predictors of cell-cycle expression were enriched for known cell-cycle regulators but also include regulators not implicated in cell-cycle regulation previously. In addition, ChIP and Deletion datasets led to the identification different subsets of TFs important for predicting cell-cycle expression. Finally, analysis of important TF-TF interaction features suggests that the GRN regulating cell cycle expression is highly interconnected and clustered around four groups of genes, two of which represent known cell-cycle regulatory complexes, while the other two contain TFs that are not known cell-cycle regulators (Ste12-Tex1 and Rap1-Hap1-Msn4), but are nonetheless important to regulating the timing of expression. Thus, not only do our models accurately reflect what is known about the regulation of the S. cerevisiae cell cycle, they can be used to discover regulatory factors which play a role in controlling expression during the cell cycle as well as other contexts with discrete temporal patterns of expression.