Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Dissertation__Durmaz.pdf (52.94 MB)

ETD Abstract Container

Abstract Header

Data Driven Approaches for Dissecting Tumor Heterogeneity

Durmaz, Arda
http://rave.ohiolink.edu/etdc/view?acc_num=case1671251436821214

Abstract Details

2023, Doctor of Philosophy, Case Western Reserve University, Nutrition.
Molecular heterogeneity in cancer has been recognized as one of the main drivers of disease relapse and drug resistance. In addition, effects of tumor-microenvironment have shown to contribute to the diversity as well. Consequently, cancer research has aimed at generating and utilizing inherently high-dimensional molecular datasets for the past decade to characterize tumors specifically with the development of `sequencing-by-synthesis' Next-Generation Sequencing (NGS) platforms. Large collections of high-dimensional multi-omics datasets exemplified by TCGA and PCAWG, 1) elaborate on the heterogeneity of cancer progression and 2) allow for increasingly complex models to be utilized. Respecting the black-box nature of machine-learning driven models, here we develop multiple strategies to leverage molecular information to delineate disease progression/mechanisms in Leukemia. Furthermore, we show the requirement of careful selection of strategies in noisy scRNA-Seq datasets in solid tumors and we propose an integrative model to investigate collateral drug-responses in a pan-cancer fashion. We present multiple strategies in two parts. First, we present a Bayesian Latent Class Analysis to incorporate molecular information in a large cohort of (n=2681) AML patients with heterogeneous characteristic and generate novel unsupervised clusters with clinical relevance. Furthermore, we utilize Autoencoder structure to develop distance-based, low dimensional clustering model to group MDS patients (n=3588) into 14 novel groups. This approach allowed us to extract relevant features otherwise difficult to capture with Bayesian strategies in noisy datasets. In the second part, we conduct a comprehensive benchmarking study to evaluate the vast repository of methods developed for scRNA-Seq analysis. We show, in contrast with the current practice, scRNA-Seq analysis is amenable to variation and results, specifically unsupervised clustering, is of qualitative nature rather than quantitative. Finally, borrowing from the power of complex neural-network based models, we develop an integrative model to capture co-varying features of gene-expression and mutation profiles of cell-lines and patient samples relevant to collateral drug response profiles.
Jacob Scott (Advisor)
178 p.
Bioinformatics
Myeloid Neoplasms, Heterogeneity, Machine-Learning, single-cell RNA-Seq

Recommended Citations

Citations

  • Durmaz, A. (2023). Data Driven Approaches for Dissecting Tumor Heterogeneity [Doctoral dissertation, Case Western Reserve University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=case1671251436821214

    APA Style (7th edition)

  • Durmaz, Arda. Data Driven Approaches for Dissecting Tumor Heterogeneity. 2023. Case Western Reserve University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=case1671251436821214.

    MLA Style (8th edition)

  • Durmaz, Arda. "Data Driven Approaches for Dissecting Tumor Heterogeneity." Doctoral dissertation, Case Western Reserve University, 2023. http://rave.ohiolink.edu/etdc/view?acc_num=case1671251436821214

    Chicago Manual of Style (17th edition)

case1671251436821214
60
© 2022, all rights reserved.
This open access ETD is published by Case Western Reserve University School of Graduate Studies and OhioLINK.