Development of Statistical Methods for Analysis of High-Dimensional Biological Data

Abstract

High-dimensional biological data have been increasingly made available for tackling complex health problems. As with any Big Data opportunities, this has led to methodological challenges for extracting relevant information from such data, particularly in settings where biologically-sensible and statistically-appropriate methodologies that are practical and effective in public health practice or healthcare delivery have not been established. This thesis aims at developing statistical methods specifically for two heath problems with high-dimensional biological data: I) A logic-regression-based genetic biomarker discovery method for environmental health, identifying the source/host of Escherichia coli using its genomic data; and II) An image analysis method for automatic tuberculosis (TB) detection in resource-limited settings, where the modern TB detection methods are not employable, using high-throughput sputum-culture images. My research has developed these methods that are aimed to be implemented in the respective fields to advance effectiveness of the public health practice.