Abstract/Project Summary Germline genetics research often relies heavily on large-scale hypothesis testing to detect associations between complex traits and their risk mutations. Traditionally, the statistical methods that we and others have developed for these settings have focused on powerful approaches that test global null hypotheses - identifying the existence of any signal in a set of individual tests. In recent years though, researchers have also increasingly emphasized novel study designs that are more suitable for another class of tests known as composite null hypothesis tests. Roughly speaking, the goal of a composite null hypothesis test is to identify the existence of multiple, as opposed to at least one, signal in a set of individual tests. However, the lack of validated variants identified by composite null analyses belies the high popularity of such studies, suggesting a lack of suitable quantitative approaches. The central goal of this proposal is to develop high-dimensional composite null hypothesis testing approaches that (a) provide interpretable results addressing the scientific question of interest and (b) offer robust performance over varied genome-wide settings. Specifically, one study design of interest is (i) pleiotropy studies. A common goal in performing pleiotropy analysis is to identify variants linked to multiple diseases simultaneously, which may, for example, suggest new indications for existing therapies. Many existing pleiotropy approaches test global nulls, which are still applicable but often less pertinent. Another related study type is (ii) mediation analysis. When testing for genetic mediation, interest lies in the simultaneous associations of a variant to a mediator and the mediator to an outcome. Thus, a composite null approach that identifies at least two associations is more suitable for this type of investigation. Additionally, (iii) replication studies are ubiquitous in genetics research and require more than one association to declare a successful replicated effect. We will develop approaches that leverage our existing empirical Bayes tools to perform composite null inference in (i)-(iii). Different models will be proposed to address unique statistical challenges such as the within-set correlation that arises in (ii) and directionality restrictions in (iii). Extensive simulation studies and real data examples will be used to illustrate (a) and (b), as well as demonstrate advantages over global null interpretations. Successful completion of this work will result in novel, interpretable, and robust strategies to perform large-scale pleiotropy, mediation, replication, and other translational genetics studies across a variety of phenotypes. We will also develop well-documented, publicly available software packages to share this methodology with the research community. Summary data from analysis of varied phenotypes will be released.