Coupled Mixed Model for joint genetic analysis of complex disorders from independently collected data sets: application to Alzheimer's disease and substance use disorder

In the last decade, Genome-wide Association studies (GWASs) have contributed to decoding the human genome by uncovering genetic variations associated with diseases. “Joint analysis”, which involves analyzing multiple independently generated GWAS data sets, has been explored extensively as follow-ups. However, most of the analyses remain in the preliminary stage. In this paper, we propose a method, called Coupled Mixed Model (CMM), that allows performing the joint analysis of GWAS on two independently collected data sets with different phenotypes, using a set of multivariate sparse mixed models. The CMM method does not require the data sets to have the same phenotypes as it aims to infer the uncollected phenotypes using statistical learning modeling. Moreover, CMM takes account of the confounders due to population stratification, family structures, and cryptic relatedness, as well as the confounders arising during data collection that frequently appear in joint genetic studies. In this work, we verify the performance of our method using simulation experiments. We also conduct a real data joint analysis of two independent data sets which were generated for investigation of genetic association for Alzheimer’s disease and substance use disorder, respectively. Our results reveal new insights into these diseases. A Python implementation of the software is available at: https://github.com/HaohanWang/CMM Author summary We present a method, namely Coupled Mixed Model, that allows user to conduct joint GWAS analysis using independently collected data sets, even when there are no shared phenotypes collected among subjects in the two data sets. Our method can automatically infer the missing phenotypes in the joint analysis using likelihood estimation given that the phenotypes are related. More importantly, joint GWAS analyses across data sets are usually considered challenging due to the existence of confounding factors including population stratification, family structure, confounders due to data collection, or cryptic relatedness. After extensive simulation experiments, we verify the performance of our method. We further apply our method to jointly analyze common genetic factors underlying both Alzheimer’s disease and substance use disorder.