Methods and Applications for Collection, Contamination Estimation, and Linkage Analysis of Large-scale Human Genotype Data

Abstract

In recent decades statistical genetics has contributed substantially to our knowledge of human health and biology. This research has many facets: from collecting data, to cleaning, to analyzing. As the scope of the scientific questions considered and the scale of the data continue to increase, these bring additional challenges to every step of the process. In this dissertation, I describe novel approaches for each of these three steps, focused on the specific problems of participant recruitment and engagement, DNA contamination estimation, and linkage analysis with large data sets. In Chapter 1, we introduce the subject of this dissertation and how it fits with other developments in the generation, analysis and interpretation of human genetic data. In Chapter 2, we describe Genes for Good, a new platform for engaging a large, diverse participant pool in genetics research through social media. We developed a Facebook application where participants can sign up, take surveys related to their health, and easily invite interested friends to join. After completing a required number of these surveys, we send participants a spit kit to collect their DNA. In a statistical analysis of 27,000 individuals from all over the United States genotyped in our study, we replicated health trends and genetic associations, showing the utility of our approach and accuracy of self-reported phenotypes we collected. In Chapter 3, we introduce VICES (Verify Intensity Contamination from Estimated Sources), a statistical method for joint estimation of DNA contamination and its sources in genotyping arrays. Genotyping array data are typically highly accurate but sensitive to mixing of DNA samples from multiple individuals before or during genotyping. VICES jointly estimates the total proportion of contaminating DNA and identify which samples it came from by regressing deviations in probe intensity for a sample being tested on the genotypes of another sample. Through analysis of array intensity and genotype data from HapMap samples and the Michigan Genomics Initiative, we show that our method reliably estimates contamination more accurately than existing methods and implicates problematic steps to guide process improvements. In Chapter 4, we propose Population Linkage, a novel approach to perform linkage analysis on genome-wide genotype data from tens of thousands of arbitrarily related individuals. Our method estimates kinship and identical-by-descent segments (IBD) between all pairs of individuals, fits them as variance components using Haseman-Elston regression, and tests for linkage. This chapter addresses how to iteratively assess evidence of linkage in large numbers of individuals across the genome, reduce repeated calculations, model relationships without pedigrees, and determine segregation of genomic segments between relatives using single-nucleotide polymorphism (SNP) genotypes. After applying our method to 6,602 individuals from the National Institute on Aging (NIA) SardiNIA study and 69,716 individuals from the Trøndelag Health Study (HUNT), we show that most of our signals overlapped with known GWAS loci and many of these could explain a greater proportion of the trait variance than the top GWAS SNP. In Chapter 5, we discuss the impact and future directions for the work presented in this dissertation. We have proposed novel approaches for gathering useful research data, checking its quality, and detecting associations in the investigation of human genetics. Also, this work serves as an example for thinking about the process of human genetic discovery from beginning to end as a whole and understanding the role of each part.