Detection of genetic factors associated with multiple correlated imaging phenotypes by a sparse regression model

Recently, more evidence of polygenicity and pleiotropy has been found in genome-wide association (GWA) studies of complex psychiatric diseases (e.g., schizophrenia), where multiple interacting genetic variants may affect multiple phenotypic traits simultaneously. In this work, we propose a new sparse collaborative group-ridge low-rank regression model (sCGRLR) to study the pleiotropic effects of a group of genetic variants on multiple imaging-derived quantitative traits (i.e., endophenotype). In the method, we enforce sparse and low-rank regularizations to reduce the number of features and then construct an effective gene or gene-set based statistic test to evaluate the significance of selected features. We show the advantage of our method with other gene-set pleiotropy analysis methods and other sparse multivariate regression methods in terms of type I error and power on simulated data. Finally, we demonstrate its application to real data analysis on the study of schizophrenia.