Statistics PhD student Manuel Garcia Acosta's defense titled "Statistical Methods for Complex Biological Data: From Omics to Infection Timing". 

This talk presents two projects that develop statistical methodology for high-dimensional omics data and infectious disease modeling.

High-throughput biological data often exhibit complex correlation structures driven by both observed covariates and latent factors. Misspecifying this dependence can bias inference and reduce the effectiveness of downstream analyses.

In this work, we develop a hypothesis testing framework for selecting among competing variance models in high-dimensional settings. By representing covariance as a linear combination of known matrices, our method identifies the components needed to capture dependence in the data. We establish theoretical guaranties and demonstrate the approach on a real-world gene expression dataset.

Respiratory syncytial virus (RSV) is a common early-life infection whose timing varies across individuals and plays an important role in subsequent respiratory outcomes. Recent work has focused on estimating the age-dependent risk of infection using population-level surveillance data on viral circulation, often in combination with scheduled testing and illness-triggered testing in cohort studies. However, complete and reliable infection data are generally not available at the population level in epidemiological studies.

In this work, we address this challenge by leveraging data from healthcare encounters related to RSV-associated illness. We extend an existing modeling framework to jointly estimate the age-dependent risk of infection and the probability that an infection results in a healthcare encounter. Our approach combines surveillance-based measures of viral circulation with flexible spline-based representations of age-dependent functions, estimated via penalized likelihood. We develop an efficient computational strategy for model fitting and demonstrate its performance through simulation studies that reflect realistic epidemiological settings.

Advisor: Dr. Chris McKennan