Electronic Theses and Dissertations
Date
2024
Document Type
Dissertation
Degree Name
Doctor of Philosophy
Department
Epidemiology
Committee Chair
Hongmei Zhang
Committee Member
Ebenezer George
Committee Member
John Holloway
Committee Member
Xichen Mou
Abstract
This dissertation explores Bayesian approaches for network analysis in the context of epigenetic modifications, specifically focusing on Deoxyribonucleic acid methylation (DNA-m) at cytosine-phosphate-guanine (CpG) sites. DNA-m is a key epigenetic modification associated with diseases and infections. Recent research suggests that analyzing joint activities across multiple CpG sites could provide more powerful insights, which can be achieved using network-based approaches, such as Gaussian networks. The first project introduces a Bayesian variable selection method for detecting singletons and reconstructing networks that incorporate singleton status. Singletons, or isolated nodes, can lead to false connections and biased interpretations. The proposed method effectively detects singletons, resulting in higher-quality network inference than methods that do not account for these nodes. The second project presents a composite likelihood-based approach to improve the scalability of an existing network differentiation method, which compares undirected Gaussian network structures under varying conditions, such as disease status or time. The proposed approach improves upon the manifest-data likelihood (MDL) method, especially for large networks, by reducing computational complexity. While effective for independent networks, this approach also shows potential for differentiating dependent networks, particularly when sample sizes are adequate. In the third project, a Bayesian approach is applied to compare directed networks generated from two dependent populations when graph ordering is unknown. Simulations indicate that the method effectively detects differentiation between networks; however, its accuracy in edge direction and exclusion of non-existent edges is limited, suggesting that it may be unsuitable for graph construction in dependent populations. This work advances network-based approaches for understanding complex dependencies in epigenetics and highlights the importance of accurate singleton detection, scalable differentiation methods, and robust statistical comparison in network analysis.
Library Comment
Dissertation or thesis originally submitted to ProQuest.
Notes
Embargoed until 11-25-2025
Recommended Citation
Adefisoye, James Olusegun, "Network Differentiation and Identification of Singletons in Gaussian Networks Under the Bayesian Framework" (2024). Electronic Theses and Dissertations. 3679.
https://digitalcommons.memphis.edu/etd/3679
Comments
Data is provided by the student.