Author ORCID Identifier
Semester
Fall
Date of Graduation
2024
Document Type
Dissertation
Degree Type
PhD
College
Eberly College of Arts and Sciences
Department
Mathematics
Committee Chair
Dr. David Klinke
Committee Co-Chair
Dr. Adam Halasz
Committee Member
Dr. Marjorie Darrah
Committee Member
Dr. Robert Mnatsakanov
Committee Member
Dr. Sijin Wen
Committee Member
Dr. Donald Adjeroh
Abstract
Bayesian network inference has revolutionized our capacity to predict causal relationships among variables in a variety of disciplines as a result of advancements in computational power. Nevertheless, the predicted causal networks are often obscured by the influence of specific computational algorithms. To address this challenge, we adopted a "wisdom of the crowds" approach and developed BaMANI (Bayesian Multi-Algorithm causal Network Inference), an ensemble learning approach that reduces the bias of individual algorithms in Bayesian causal network inference. This thesis establishes the theoretical foundations of our methodology, provides a comprehensive implementation of BaMANI as a software tool, and illustrates its utilization in a human breast cancer study.
We utilized BaMANI in order to investigate the network of interest \textcolor{black}{related to a protein that is} secreted by malignant cells \textcolor{black}{and that could} potentially mediate immunosuppression. \textcolor{black}{This protein is thought to} influence the dynamics of the tissue microenvironment \textcolor{black}{by altering} interactions among various cell types, thereby changing the composition and functionality of cells within tissues during oncogenesis. Our objective was to elucidate the causal dynamics between a variety of cell types and a protein \textcolor{black}{of interest}, with a particular emphasis on the impact of the protein \textcolor{black}{of interest} on cellular composition. The analysis was performed on a dataset obtained from the Cancer Genome Atlas (TCGA), which included transcriptomic profiling of 582 breast cancer tissue samples. The dataset includes various cell types and features, involving immune cells (e.g., CD4 and CD8 T cells, Neutrophils, Macrophages), stromal cells (e.g., Endothelial cells, Cancer-Associated Fibroblasts), and other cell types (e.g., B cells, Epithelial, Mesenchymal cells). Additionally, the dataset includes features such as proliferation, CCN4, and specific \textcolor{black}{macrophage subtypes M0, M1, and M2.}
The findings indicate that BaMANI effectively quantifies and identifies the impact of specific proteins on the structure of the tumor microenvironment, and demonstrates how this protein affect cell quantities and configure the causal network, \textcolor{black}{once validated, this knowledge can advance} our understanding of cancer biology and contributes to the development of more effective therapeutic strategies. \textcolor{black}{To validate the biological relevance of the predictions made by BaMANI, laboratory in vivo verification tests were conducted as part of other \textcolor{black}{NIH/NSF research projects \citep{Fernandez_Klinke_EMBO_rep_2022,Pirkey_Klinke_CMB_2023} in the Lab} to determine whether inhibiting the production of these components in tumor cells enhances the immune system’s ability to control tumor growth}. BaMANI's robustness is also demonstrated by the use of performance measurements, which include comparative network analysis using multiple individual algorithms and arc strength analysis. This study not only demonstrates the capability of BaMANI to provide deeper understanding of the cellular dynamics of cancer, but also highlights its utility in broader contexts where comprehension of underlying causal relationships is crucial in a dynamic system.
Recommended Citation
Latifizadeh, Habibolla, "Developing a data-driven method for inferring heterocellular networks in cancer" (2024). Graduate Theses, Dissertations, and Problem Reports. 12716.
https://researchrepository.wvu.edu/etd/12716
