Abstract:
The subject of this thesis is the inference of regulatory networks from mRNA transcript abundance data, often described as gene expression data, and its application to problems in human biology. A unified software framework is developed for the implementation and operation of network inference algorithms. This framework, the Network Inference Processing Framework, has been designed to integrate inference algorithms produced across different technologies by different research groups. Within this software framework, a number of published network inference methods are implemented, including Ml ΚΑΝΑ (Method for Inferring Kinetics and Network Architecture), a recently developed application of an iterative linear regression method for network inference. Other algorithms implemented within the framework are ARACNE, a mutual-information-based method, BANJO, a Bayesian statistical method, and standard linear correlation. The behaviour of each method is investigated in detail, with the goal of setting parameters for subsequent work. New methods to identify and visualize significant or meaningful features in gene networks are developed to highlight the similarities and differences between gene networks produced by the different methods implemented within the Network Inference Processing Framework. Using the network inference methods and the visualization tools developed in the thesis, the action of the Rel/NF-κΒ family of transcription factors in human umbilical vein endothelial cells (HUVEC). A new hypothesis is proposed to explain the manifestation of interactions up- and downstream of Rel/NF-κΒ activity in the data. Finally, the tools developed are applied to investigate the biology of malignant melanoma. The creation of a new microarray dataset representative of cell-cycle events in melanoma is described, and this data is combined with published clinical data from patient tumours to identify previously unknown master regulators associated with differences in survival of melanoma patients.