A New Canonical Correlation Association Measure for Multivariate Analysis with Massive Categorical Data

• Activity Number: 359
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Hongyan Chen*+ and Laura Jean Bierut and Wei Zhu
• Companies: Stony Brook University and Washington University at St. Louis and The State University of New York at Stony Brook
• Keywords: GWAS ; SNPs ; categorical ; canonical correlation ; association measure

Abstract

Fueled by the modern genome-wide association studies (GWAS), there is a growing need for novel multivariate analysis methods for massive categorical data. The goal of GWAS is to investigate the relationship between phenotypes and genotypes, which can be determined by over a million single nucleotide polymorphisms (SNPs). Such high dimensionality demands more efficient multivariate analysis tools such as cluster analysis (CA) and partial correlation network analysis (PCNA). However, both of them are mainly developed for the numeric data. An urgent task is to customize them for categorical one by developing suitable association/distance measure. In this paper, we first examine the performance of several existing measures between SNPs. Subsequently, by treating SNPs as categorical variables, we propose a novel (partial) canonical correlation association measure that has significant advantages. In applying this new association measure to COGEND, a small GWAS study on nicotine-addiction, We discovered that the CA achieved more accurate chromosomal separation and the hub-SNPs identified by the PCNA yielded higher prediction accuracy for nicotine-addiction status.