Explicit Modeling of RNA Stability Improves Large-Scale Inference of Transcription Regulation

Inference of eukaryotic transcription regulatory networks remains challenging due to the large number of regu-lators, combinatorial interactions, and redundant pathways. Even in the model system Saccharomyces cerevisiae, inference has performed poorly. Most existing inference algorithms ignore crucial regulatory components, like RNA stability and post-transcriptional modulation of regulators. Here we demonstrate that explicitly modeling tran-scription factor activity and RNA half-lives during inference of a genome-wide transcription regulatory network in yeast not only advances prediction performance, but also produces new insights into gene-and condition-specific variation of RNA stability. We curated a high quality gold standard reference network that we use for priors on network structure and model validation. We incorporate variation of RNA half-lives into the Inferelator inference framework, and show improved performance over previously described algorithms and over implementations of the algorithm that do not model RNA degradation. We recapitulate known condition-and gene-specific trends in RNA half-lives, and make new predictions about RNA half-lives that are confirmed by experimental data.