Biological network models have become standard tools for genome-wide analysis of both cancer disease processes and healthy differentiation from stem cells. In this thesis, I address a method for evaluating network models in terms of their ability to predict held out expression data given information about other genes in the same network. I apply this test to several extensions of our pathway database to demonstrate the transcriptional modeling utility of reverse phase protein array data, natural language processed literature, and transcription factor target predictions in stem cells and differentiated tissue. I then explore the addition of one new type of network data; co-localization in DNA domains. Preexisting functional data is not the only source of networks. I also elucidate a method to integrate prior biological knowledge with time series observations of reverse phase protein arrays to infer causal relationships between phosphorylation events on proteins using a novel extension of L1 penalized Granger Causality and a heat diffusion graph prior.