Integrative statistical methods for the genomic analysis of immune-mediated disease

Genome-wide association studies (GWAS) have proved to be a successful method in cataloguing loci influencing thousands of complex human disease phenotypes. However, elucidating the causal mechanisms underlying such associations has proved challenging due to the regulatory nature of the majority of signals. In Chapters 2 and 3, I hypothesised that promoter-capture Hi-C (PCHi-C) data might have utility in physically linking disease-associated regulatory variants to their target genes, in a tissue-specific manner. To examine the genome-wide enrichment of GWAS summary statistics within PCHi-C chromatin contact maps I developed a novel statistical method, blockshifter. I applied \textit{blockshifter} to a compendium of GWAS summary statistics for 31 traits and PCHi-C data across 17 primary blood tissues, and found convincing evidence for the enrichment of immune-mediated disease (IMD) GWAS signals in lymphocyte-specific chromatin interactions, providing support for the hypothesis. Taking a more gene-centric approach I developed `COGS', a novel method for integrating GWAS and PCHi-C to prioritise specific causal variants, genes and cellular contexts for functional follow up. With a focus on IMD, I prioritised tissue-context specific interactions in CD4$^{+}$ T cells linking putative causal variants for type 1 diabetes, to the promoter of IL2RA. The effect of these variants on IL2RA expression was subsequently validated by allele-specific expression, by a collaborator, supporting the approach. In Chapter 4, I hypothesised that summary statistics from multiple, well-powered GWAS of related diseases might be exploited to provide insight into rarer related diseases or disease subtypes. To investigate this I developed a PCA based framework to generate a lower-dimensional basis, summarising input GWAS traits. I constructed such a basis from ten IMD GWAS studies, excluding variants in the HLA region, and projected on summary GWAS data from multiple sources in order to characterise individual principal components (PCs). By projecting on both summary and individual-level genotype data for juvenile idiopathic disease subtypes, I was able to show that a single PC was able to discriminate enthesitis-related and systemic forms of the disease from other subtypes