Discovery of multivariate phenotypes using association rule mining and their application to genome-wide association studies

Genome-wide association studies (GWAS) have served crucial roles in investigating disease susceptible loci for single traits. On the other hand, the GWAS have been limited in measuring genetic risk factors for multivariate phenotypes from pleiotropic genetic effects of genetic loci. This work reports a data mining approach to discover patterns of multivariate phenotypes expressed as association rules, and present an analytical scheme for GWAS of those multivariate phenotypes as defining new phenotypes. We identified 13 SNPs for four genes (CSMD1, NFE2L1, CBX1, and SKAP1) associated with low levels of low density lipoprotein cholesterol (LDL-C ≤ 100 mg/dl) and high levels of triglycerides (TG ≥ 180 mg/dl) as a multivariate phenotype. Compared with a traditional approach to GWAS, the use of discovered multivariate phenotypes can be advantageous in identifying genetic risk factors, accounting for pleiotropic genetic effects when the multivariate phenotypes have a common etiologic pathway.