Methods for integrative genome-scale inference of transcriptional gene regulation

In this thesis I discuss approaches for integrating prior knowledge into methods for gene regulatory network inference, and their application to B. subtilis and O. sativa. I also examine the inter-species translation of gene expression from rat to human.;In Chapter 1 I and co-authors present two methods for incorporating additional knowledge to constrain global regulatory network inference by adding priors on the network structure. We show that both methods are remarkably tolerant to error in the priors, and that the inclusion of prior knowledge significantly improves the quality of inferred networks without damaging our ability to learn new interactions.;Chapter 2 proposes methods to use rat transcription data from specific stimuli to predict human gene set and pathway activity under the same perturbations. The methods were submissions to the IMPROVER Species Translation Challenge and evaluated using training and test set data. An important outcome was the identification of human and rat gene pairs thought to be under similar regulatory control for the given type of cells and stimuli used in the experiments. Our gene pairs predicted inter-species differential gene expression better than sequence based orthologs and were significantly overlapping between the two methods, yet they were divergent from sequence based orthologs extracted from HGNC database.;Finally, in Chapter 3 I show how we can integrate network component analysis (NCA) into our existing inference approach. We use NCA and prior networks to estimate transcription factor activities, which greatly improve the accuracy of inferred interactions. The existence of a high-quality gold standard and knock-out experiments in the B. subtilis project allowed us to evaluate our new method showing unprecedented accuracy. With the rice project we could show that we can take the same approach and apply it to a more complex and less well studied organism by generating a network prior from chromatin accessibility data and transcription factor biding motifs.