Graduation Year

2026

Document Type

Thesis

Degree

M.S.

Degree Name

Master of Science (M.S.)

Degree Granting Department

Pharmacy

Major Professor

Feng Cheng, Ph.D.

Committee Member

Vijaykumar Sutariya, M.Pharm, Ph.D., R.Ph

Committee Member

Sheeba Varghese Gupta, M.Pharma, Ph.D.

Keywords

EHR, FAERS, Hemolysis, Machine Learning, Pharmacovigilance, Transcriptomics

Abstract

This thesis applied integrative computational approaches to two major areas of drug evaluation and discovery: cardiovascular safety assessment and peptide hemolytic activity prediction. First, pharmacovigilance data from the FDA Adverse Event Reporting System (FAERS), electronic health records (EHRs) from TriNetX, and transcriptomic analyses were combined to evaluate anesthetic-associated cardiac rhythm adverse drug events. Among propofol, ketamine, dexmedetomidine, and midazolam, dexmedetomidine showed the strongest association with bradycardia across both FAERS and propensity score-matched EHR analyses. Stratified analyses further suggested substantially higher bradycardia prevalence among older patients and a modestly increased risk among male patients. Drug-drug interaction analyses identified risperidone and albuterol as co-medications associated with increased bradycardia risk during dexmedetomidine exposure. Transcriptomic analyses provided supportive biological context, including downregulation of cardiac muscle contraction-related genes and identification of miR-26a-5p and miR-30c-5p as candidate regulators of heart-rate-related pathways. Second, a similar real-world data and transcriptomic framework was applied to compare cardiovascular adverse events associated with three COVID-19 therapies, remdesivir, Paxlovid, and REGEN-COV. Remdesivir showed the strongest and most consistent cardiovascular safety signals, particularly for bradycardia, hypotension, and cardiac arrest. Age-stratified analyses indicated distinct age-associated patterns, while transcriptomic analysis of remdesivir-treated cardiomyocytes identified enrichment of pathways related to cGMP-PKG signaling, calcium signaling, cardiac muscle contraction, and cardiomyopathy. Third, this thesis developed an interpretable XGBoost-based model to predict peptide hemolytic activity using sequence-derived physicochemical and mechanism-based features. The model achieved performance comparable to existing predictors while providing improved interpretability. Feature importance and correlation analyses highlighted hydrophobic patch count as a key sequence-level feature associated with hemolytic activity, suggesting that local hydrophobic organization may be more informative than global hydrophobicity alone. Overall, this thesis demonstrates the value of integrating real-world data, transcriptomics, and interpretable machine learning to identify clinically relevant safety signals and guide safer therapeutic design.

Download

Share

COinS