Integrative statistical methods for human biology, from biobanks to perturbation atlases

Abstract

Since the human genome project, many new types of rich biological data have empowered new insight into basic biological science and human disease. Among these are genetic association studies, which characterize many genetic variants in large numbers of individuals, and perturbation atlases, which profile how genetic perturbations shape the expression of thousands of transcripts. However, statistical methods for answering basic questions about these data (How much signal is in each experiment? What is the relationship between experiments? How many discoveries will we make at massive sample size?) lag behind. This thesis presents a suite of novel statistical approaches for the analysis of rich biological data. Our analyses clarify several integrative questions about human biology, ranging from the contribution of rare genetic variation to common diseases, to the consistency of gene perturbation effects across cell types, among others. Looking forward, these methods contribute to the creation of a harmonized analysis toolbox to facilitate replicable characterization and comparison of high-throughput biological experiments.