The survival of any organismis determined by the mechanisms triggered in response to the inputs received. Underlying mechanisms are described by graphical networks that can be inferred from different types of data such as microarrays. Deriving robust and reliable networks can be complicated due to the microarray structure of the data characterized by a discrepancy between the number of genes and samples of several orders of magnitude, bias and noise. Researchers overcome this problem by integrating independent data together and deriving the common mechanisms through consensus network analysis. Different conditions generate different inputs to the organism which reacts triggering different mechanisms with similarities and differences. A lot of effort has been spent into identifying the commonalities under different conditions. Highlighting similarities may overshadow the differences which often identify the main characteristics of the triggered mechanisms. In this thesis we introduce the concept of study-specific mechanism. We develop a pipeline to semiautomatically identify study-specific networks called unique-networks through a combination of consensus approach, graphical similarities and network analysis. The main pipeline called UNIP (Unique Networks Identification Pipeline) takes a set of independent studies, builds gene regulatory networks for each of them, calculates an adaptation of the sensitivity measure based on the networks graphical similarities, applies clustering to group the studies who generate the most similar networks into study-clusters and derives the consensus networks. Once each study-cluster is associated with a consensus-network, we identify the links that appear only in the consensus network under consideration but not in the others (unique-connections). Considering the genes involved in the unique-connections we build Bayesian networks to derive the unique-networks. Finally, we exploit the inference tool to calculate each gene prediction-accuracy across all studies to further refine the unique-networks. Biological validation through different software and the literature are explored to validate our method. UNIP is first applied to a set of synthetic data perturbed with different levels of noise to study the performance and verify its reliability. Then, wheat under stress conditions and different types of cancer are explored. Finally, we develop a user-friendly interface to combine the set of studies by using AND and NOT logic operators. Based on the findings, UNIP is a robust and reliable method to analyse large sets of transcriptomic data. It easily detects the main complex relationships between transcriptional expression of genes specific for different conditions and also highlights structures and nodes that could be potential targets for further research.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:642441 |
| Date | January 2014 |
| Creators | Bo, Valeria |
| Contributors | Tucker, A.; Swift, S. |
| Publisher | Brunel University |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://bura.brunel.ac.uk/handle/2438/10390 |
Page generated in 0.0025 seconds