Abstract
Pleiotropy, the phenomenon where a genetic region confers risk to multiple traits, is widely observed, even among seemingly unrelated traits. Knowledge of pleiotropy can improve understanding of biological mechanisms of diseases/traits, and can potentially guide identification of molecular targets or help predict side effects in drug development. However, statistical approaches for identifying pleiotropy genome-wide are limited, particularly for two correlated traits or case-control traits with unknown sample overlap or for disease traits from family studies. We proposed PLACO+, an improved version of our pleiotropic analysis under composite null hypothesis method based on GWAS summary statistics from two traits. PLACO+ uses an inflated variance model to allow for fractions of variants to be associated with none or only one trait under the null. It is genome-wide scalable, where analytical p value is computed as a weighted sum of extreme tail probabilities of bivariate normal product distribution. Simulations for both population-based and family-based designs demonstrate well-calibrated type I errors at stringent levels and substantially improved power of PLACO+ over conventional approaches. We illustrate PLACO+ on inflammatory bowel disease subtypes with shared controls and on correlated lipid traits with unknown sample overlap. In particular, PLACO+ revealed pleiotropic regions between triglyceride and high-density lipoprotein levels that conventional approaches missed and all of which were replicated in a larger GWAS of these lipid traits. This further demonstrates the utility of PLACO+ in discovering genetic associations of traits with modest sample sizes by leveraging information from another correlated trait.