Background: Although pleiotropy, which occurs when a genetic variant affects at least two traits, is thought to play a central role in the genetic architecture of complex traits and diseases, it is a poorly understood mechanism. Our objective is to build a genome-wide map of pleiotropic variants of the human genome.

Methods: We have developed PleioVar, an algorithm to label pleiotropic variants in the human genome. PleioVar derives variant pleiotropic labels from integrative Mendelian Randomization methods such as LHC-MR or MR-CUE. The objective is to discern the biological mechanisms leading to observed pleiotropy. We propose 5 types of pleiotropy (direct, confounding-mediated, traits-mediated, linkage disequilibrium and serendipitous) and model these pleiotropies with Gaussian mixture models, at the level of variants. We selected highly heritable traits and diseases from UK Biobank and built a complex network GWAS simulation framework based on these traits to assess the performance of PleioVar. Then, using GWAS summary statistics from UK Biobank, we used PleioVar to detect pleiotropic variants of these same traits.

Results: On simulations, PleioVar succeeded to predict variant pleiotropy with high accuracy. Simulated genetic variants were correctly clustered as displaying the different types of pleiotropy. Using GWAS summary statistics from highly heritable traits and diseases from the UK Biobank, pleiotropic labels of genetic variants were obtained from PleioVar.

Conclusion: This tool, designed to map genetic variant pleiotropy, could help better understand the human genetic architecture and characteristics of complex traits and diseases.