Interpretable and reliable statistical models for biomedicine

Abstract

This dissertation presents a collection of statistical tools to analyze modern biomedical datasets, which have been transformed by developments in high-throughput and high- content biology. Due to rapid growth in both scale and complexity of these datasets, there is a need for new inference procedures that combine both statistical and computational perspectives. Our results have been organized according to two major themes. In Part One, we use modern sequencing and gene-editing technologies to understand immunological and developmental processes. In Part Two, we study the accuracy and reliability of non-parametric machine learning algorithms, motivated by their growing use in medicine and healthcare.