Investigating human disease mechanisms through population genetics and experimental genetic screens

Abstract

In this dissertation, I develop new computational and/or experimental methods to elucidate human disease mechanisms from two different data modalities: (1) genotypes with matched phenotypes and gene expression profiles collected from large human populations, and (2) experimental genetic screens that perturb cells using CRISPR and read out full gene expression profiles using single-cell RNA-sequencing (i.e. Perturb-seq). In Chapter 1, I introduce a new method called “mediated expression score regression” (MESC) to estimate a quantity called heritability mediated by gene expression, which can be conceptualized as the total genetic effect on a phenotype in a human population that “flows through” gene expression measurements. In Chapter 2, I apply MESC to estimate the proportion of heritability for 42 phenotypes (ranging from height to diabetes to schizophrenia) mediated by gene expression measurements from the largest available datasets, showing that only a small proportion of heritability is mediated by gene expression measurements. In Chapter 3, I introduce new experimental and computational methods to run Perturb-seq more efficiently based on “random composite perturbations,” which are inspired by theoretical results from compressed sensing. I show that our new approach to running Perturb-seq (which we call “compressed Perturb-seq”) achieves an order of magnitude cost reduction over existing approaches while maintaining the same accuracy. In Chapter 4, I describe the biological insights about the innate immune response gained from testing compressed Perturb-seq in a human macrophage cell line. In Chapter 5, I describe results from integrating our compressed Perturb-seq screen with population-genetic data, showing that gene sets generated from Perturb-seq data are highly enriched for the heritability of immune diseases, but that Perturb-seq data shows little concordance with expression quantitative trait loci (eQTLs) despite their conceptual similarities.