Date Approved

6-30-2025

Embargo Period

6-30-2025

Document Type

Dissertation

Degree Name

Ph.D. Biological and Biomedical Sciences

Department

Biological and Biomedical Sciences

College

College of Science & Mathematics

Advisor

Hao Zhu, Ph.D

Committee Member 1

Erik Hoy

Committee Member 2

Thomas Keck, Ph.D.

Committee Member 3

Zhihong Wang, Ph.D.

Committee Member 4

Lauren M. Aleksunes, Pharm. D., Ph.D., D.A.B.T.

Keywords

Adverse outcome pathway;Artificial intelligence applications;Computational toxicology;Mechanistic modeling;New approach methodologies;Quantitative structure-activity relationship modeling

Disciplines

Life Sciences | Pharmacology, Toxicology and Environmental Health | Toxicology

Abstract

Regulatory toxicology depends primarily on animal testing, which limits throughput, mechanistic interpretability, and applicability to human-relevant endpoints. This dissertation presents computational approaches that integrate publicly available bioactivity data with machine learning to support chemical safety evaluation. Oral carcinogenicity was modeled using curated chemicals from the US Environmental Protection Agency’s Integrated Risk Information System (IRIS) and activity data from over 1,900 PubChem assays. Quantitative structure-activity relationship (QSAR) models trained on statistically prioritized assays demonstrated strong predictive performance and were applied to rank the carcinogenic potential of more than 16,000 compounds. Hepatotoxicity was evaluated using a hybrid modeling approach that combined structural alerts with mitochondrial membrane potential (MMP) data from high-throughput screening (HTS), improving classification accuracy (CCR = 0.80) and experimental validation in HepG2 cells (CCR = 0.79). A hierarchical modeling framework was developed to organize HTS concentration-response data across biological levels, mapping 455 assays to 216 protein targets and 103 pathways. This model enabled predictions for five in vivo toxicity endpoints and incorporated toxicokinetic modeling to support exposure-informed risk evaluation. Collectively, these approaches advance scalable, mechanism-driven models for compound prioritization and chemical safety assessments.

Download

Included in

Toxicology Commons

Share

COinS