Genome-wide Marginal Epistatic Association Mapping in Case-Control Studies

Abstract

Epistasis, commonly defined as the interaction between genetic loci, is an important contributor to the genetic architecture underlying many complex traits and common diseases. Most existing epistatic mapping methods in genome-wide association studies explicitly search over all pairwise or higher-order interactions. However, due to the potentially large search space and the resulting multiple testing burden, these conventional approaches often suffer from heavy computational cost and low statistical power. A recently proposed attractive alternative for mapping epistasis focuses instead on detecting marginal epistasis, which is defined as the combined pairwise interaction effects between a given variant and all other variants. By searching for marginal epistatic effects, one can identify genetic variants that are involved in epistasis without the need to identify the exact partners with which the variants interact -thus, potentially alleviating much of the statistical and computational burden associated with conventional epistatic mapping procedures. However, previous marginal epistatic mapping methods are based on quantitative trait models. As we will show here, these lack statistical power in case-control studies. Here, we develop a liability threshold mixed model that extends marginal epistatic mapping to case-control studies. Our method properly accounts for case-control ascertainment and the binary nature of casecontrol data. We refer to this method as the liability threshold marginal epistasis test (LT-MAPIT). With simulations, we illustrate the benefits of LT-MAPIT in terms of providing effective type I error control, and being more powerful than both existing marginal epistatic mapping methods and conventional explicit search-based approaches in case-control data. We finally apply LT-MAPIT to identify both marginal and pairwise epistasis in seven complex diseases from the Wellcome Trust Case Control Consortium (WTCCC) 1 study.

Uncertainty Quantification in Variable Selection for Genetic Fine-Mapping using Bayesian Neural Networks

“…When disease prevalence is very low (e.g., k = 1%), cases are assumed to come from "tail" of the distribution. In this scenario, statistical models are generally better powered [27]. As the prevalence k becomes greater, such that the liability threshold moves from the tails to the center of the distribution, it will become harder for a classifier to distinguish cases from This also results in lower power for variable selection.…”

Pathway Analysis within Multiple Human Ancestries Reveals Novel Signals for Epistasis in Complex Traits

“…There are a few important takeaways from this formulation of MAPIT-R. First, the term m l effectively represents the polygenic background of all variants except for those that have been annotated for the l-th region of interest. Second, and most importantly, the term g l is the main focus of the model and represents the marginal epistatic effect of the region R l [43,44]. It is important to note that each component of the model will change with every new region that is considered.…”

“…In contrast, the pathways with significant marginal epistatic effects identified in both the African and Chinese subgroups are pathways related to the immune system and contain multiple HLA loci (e.g., HLA-DRA, HLA-DRB1, HLA-A, HLA-B ) ( Supplementary Tables 5 and 6). These results are unsurprising since it is well known that the MHC region holds significant clinical relevance in complex traits [44,103,104,120]; however, more recent work has also suggested that Han Chinese genomes may be particularly enriched for interactions involving HLA loci [121].…”

Variable prioritization in nonlinear black box methods: A genetic association case study

“…For all synthetic demonstrations and assessments, we consider a simulation design that is often used to explore the utility of statistical methods across different genetic architectures underlying complex phenotypic traits [18,19,61]. First, we assume that all of the observed genetic effects explain a fixed proportion of the total phenotypic variance.…”

“…For example, understanding how statistical epistasis between genes (i.e. the polynomial terms of the variables in the genotype matrix) influence the architecture of traits and variation in phenotypes is of great interest in genetics applications [12][13][14][15][16][17][18][19]. However, despite studies that have detected "pervasive epistasis" in many model organisms [20] and improved genomic selection (i.e.…”