Date of Award

2026-08-01

Degree Name

Doctor of Philosophy

Department

Computational Science

Advisor(s)

Jonathon Mohl

Second Advisor

Enrique Ramos

Abstract

Cancer prognosis is shaped by complex molecular and clinical factors that challenge survival modeling using high-dimensional genomic data. Transcriptomic profiling provides a means to characterize molecular patterns associated with disease aggressiveness, yet the behavior of transcriptome-informed predictive models remains sensitive to endpoint choice, cohort structure, and demographic stratification. This dissertation presents a computational framework for transcriptomic analysis and survival modeling across multiple cancer types with explicit race-aware design. Using bulk RNA-sequencing data from The Cancer Genome Atlas, transcriptomic analyses were performed for breast invasive carcinoma (BRCA), lung adenocarcinoma (LUAD), prostate adenocarcinoma (PRAD), and liver hepatocellular carcinoma (LIHC). Differential expression and functional enrichment analyses characterized prognostically relevant transcriptomic structure, while predictive survival modeling using LASSO-regularized Cox regression generated transcriptome-informed risk scores under multiple validation strategies. Results show that aggressive disease is associated with coordinated transcriptomic programs related to proliferation, DNA damage response, metabolism, and immune processes, with detectability varying by cancer context and data structure. Predictive modeling behavior was strongly influenced by event sparsity and sample size imbalance, particularly in underrepresented Black cohorts, where instability and attenuated survival separation reflected data constraints rather than model failure. Endpoint selection further shaped feasibility, with recurrence-based modeling required for indolent cancers such as prostate adenocarcinoma. This work contributes a reproducible, race-aware computational pipeline that emphasizes stability, transparency, and disciplined interpretation for transcriptomic survival modeling in precision oncology.

Language

en

Provenance

Received from ProQuest

File Size

354 p.

File Format

application/pdf

Rights Holder

Arnav Joshi

Download

Share

COinS