Generalization of kernel machine methods for association testing of multi-omics data

Abstract

Over the past couple of decades, genome-wide association studies (GWASs) have successfully identified thousands of loci associated with complex traits and diseases in humans. Despite the immense success of these statistical tools, post-GWAS, we are often left underwhelmed by findings that are difficult to interpret or fail to to lead to causal mechanisms and deeper understanding of trait etiology. Studies utilizing omics, including transcriptomics, proteomics, metabolomics, etc, are gaining popularity, and, used in conjunction with genomics, may aid in providing insight into complex trait etiology and disease pathogenesis. To fully harness the availability of multi-omics data types, we propose to jointly evaluate, at the gene or pathway level, the cumulative effect of all data types simultaneously. We perform these analyses using the kernel machine regression (KMR) testing framework. Within this context, we propose three projects. For project one, we extend an existing KMR testing method to accommodate joint association testing of two data types with a trait of interest in correlated samples. For project two, we generalize existing KMR testing methods to allow for joint association testing of as many data types as desired against a trait of interest in correlated samples. Finally in project three, we propose a pseudo-permutation approach to association testing of an omics data type with a trait in correlated samples for studies with small sample sizes. These statistical tools facilitate analysis of complex multi-omics studies that are applicable to a broad range of studies with correlated samples, including family-based studies with extensive relatedness and studies in ancestrally diverse populations.