Sonny Min
Title: Statistical and Machine Learning Methods for Biological Aging
Date: April 13th, 2026
Time: 10am
Location: LIB 2020 & Zoom
Supervised by: Lloyd Elliott & Angela Brooks-Wilson
Abstract: Aging is a complex biological process that leads to a decline in how well human bodies function. Although chronological age is the most common way to measure this process, it does not explain why some people stay healthy while others develop disease sooner. This dissertation focuses on computational methods to identify biomarkers of biological aging. By using data from the Canadian Longitudinal Study on Aging, we combined statistics and machine learning, and causal inference techniques to find physiological sweet spots, which are the specific physiological ranges for which health is most resilient. The research consists of three main projects. In the first project, we built a BTHS (Bayesian testing for heteroskedasticity and sweet spots) framework to reveal blood biomarkers related to frailty. We found that explicitly accounting for the heteroskedasticity helps us more accurately identify the optimal blood component ranges where the frailty index is the lowest (i.e., healthiest). This method identified nine blood components with sweet spots relevant to health. In the second project, we created a framework called CMAPLE (Causal mediation analysis of pathways linking exposures) to study how genetic variants affect the risk of age-related diseases through metabolites. We identified 190 causal links across six major diseases including Alzheimer鈥檚 disease and diabetes. In the final project, we developed a tabular deep learning-based tool called Q-FSNet (Quantile feature selection network), designed to analyze high-dimensional metabolomic data and identify metabolites that possess optimal physiological ranges. This model discovered 25 metabolites with sweet spots that are related to slower biological aging. Together, these computational frameworks offer a methodology for identifying physiological drivers of biological aging, providing high-priority candidates for further mechanistic investigations and clinical studies to validate these biomarkers as potential targets for age-delaying interventions.