DOI

https://doi.org/10.25772/QPZS-YJ59

Author ORCID Identifier

https://orcid.org/0009-0006-8667-9609

Defense Date

2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Biostatistics

First Advisor

Dipankar Bandyopadhyay, Ph.D.

Second Advisor

Chenlu Ke, Ph.D.

Abstract

Recent advances in sequencing technologies have enabled the collection of extensive genome wide data, significantly enhancing the diagnosis and prognosis of head and neck cancer. Identifying predictive markers for survival time is crucial for developing prognostic systems and understanding the molecular drivers of cancer progression. Moreover, given the increasing recognition of HPV-associated head and neck (H&N) cancers, understanding its role is vital for prevention and guiding treatment decisions. In an attempt to address these, this dissertation develops model-free feature screening procedures for ultra-high dimensional right censored data, capable of capturing important features, both (a) marginally and (b) conditionally (say, on HPV status), while preserving robustness against unknown censoring mechanisms. Specifically, the proposed two-stage approach initially selects significant features using nonparametric reproducing-kernel-based ANOVA statistics, and then refines them under directional false discovery rate (FDR) control through a unified knockoff procedure. Theoretical properties, including sure screening, and rank consistency were studied. The finite sample properties of the proposed method, and the novelty in light of existing alternatives were explored through simulation studies. The methodology was illustrated via application to the right-censored H&N cancer survival data derived from The Cancer Genome Atlas, and validated on a similar dataset from the Gene Expression Omnibus database. The R package DSFDRC available in GitHub implements the proposed methodology.

Rights

© The Author

Is Part Of

VCU University Archives

Is Part Of

VCU Theses and Dissertations

Date of Submission

5-28-2024

Available for download on Sunday, May 27, 2029

Share

COinS