Reconstruction of gene regulatory networks using biological domain knowledge

"A thesis submitted in fulfillment of the requirements for the degree of Doctor of Philosophy""November 2012"Bibliography: pages173-188.1. Introduction -- 2. Eukaryotic gene transcription -- 3. Current approaches to modeling gene regulatory networks -- 4. Phase-specific regulation in the yeast cell cycle -- 5. Co-regulation of co-expressed genes -- 6. Transcriptional regulation from multi-source data -- 7. Performance analysis of the GRN models -- 8. Scalability analysis -- 9. Conclusion and future work.High throughput technologies such as microarrays generate an enormous amount of genomic data at the cellular level. The computational reconstruction of gene regulatory networks (GRN) from this abundance of data has become a major area of research in systems and computational biology. However, the reconstruction task suffers from two major challenges: the excessive computational complexity and the low accuracy of the estimated networks. Literature of related works suggests the utilization of domain knowledge in addressing these challenges. The main interest of this thesis is to study the effectiveness of incorporating biological knowledge and other sources of biological data in the computational reconstruction of GRN. -- The thesis starts with the identification of several key features of gene regulation that are used by the transcriptional regulators and employs that knowledge to restrict the number of possible regulators for each gene. We choose Dynamic Bayesian Network for the computational reconstruction of the GRN. The thesis then explores the co-regulation of genes and the potency of integrating multiple sources of biological data in the reconstruction task. Through the analysis of both real and synthetic data, this thesis also quantifies to what extent the computation time and reconstruction accuracy of the model has been improved. -- The comprehensive performance and scalability analysis of various GRN models demonstrate that the employment of biological features can convincingly reduce the computational complexity of the model. Moreover, the integration of other sources of biological data makes the model computationally efficient and estimates networks with improved accuracy. Most importantly, such integration results in a scalable model; that is, the model estimates networks including thousand genes while preserving its level of performance.Mode of access: World Wide Web.1 online resource (xxi, 188 pages) illustrations (some colour