Inferring genetic regulatory network structure: Integrative analysis of genome-scale data

With the aim of uncovering regulatory relationships that underly biological processes, we constructed a framework of computational tools and techniques to relate disparate genome-scale data within and across datasets. Using these tools we focus on the yeast cell cycle and the transcriptional network driving the transition into and out of G1. Through integrative analysis of genome-scale datasets we were able to recover many of the previously known transcriptional regulatory connections within the yeast cell cycle. We also found several novel hypothetical connections yet to be experimentally validated.Much of the analysis of large-scale gene expression data has relied heavily on the application of clustering algorithms to identify sets of co-expressed genes (clusters). In chapter 2 we introduce several new techniques for comparing and evaluating microarray data, specifically focusing on clustering results. We discuss the need for quantitative methods for evaluating clustering methods, and discuss the application of comparative analysis of clustering results.Remarkably, our analysis shows the results from any clustering algorithm are quite sensitive to slight perturbations to the data. Yet, the underlying structure revealed by most clustering algorithms remains fairly stable. These findings have a pragmatic impact on how clustering results should be interpreted and used. Chapter 3 uses the tools introduced in chapter 2 and performs a systematic comparison of the influence of noise on the stability and reliability of clustering results.In chapter 4 we demonstrate the use of artificial neural networks (ANNs) to infer regulatory networks by combining expression data and protein:DNA binding data. We then compare these regulatory relationships to the presence of transcription factor binding sites. We also note evolutionary stability in some of the components of this network by comparing results to other species of yeast.