Comparison of Association Methods for Genetic Association Studies

• Activity Number: 188
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Olivia Bagley*+ and Allison Huber and Rachel Beckner and Wesley Stewart and David Reif and Alison Motsinger-Reif
• Companies: North Carolina State University and Meredith College and Meredith College and North Carolina State University and North Carolina State University and North Carolina State University
• Keywords: statistical genetics ; Multifactor Dimensionality Reduction ; LASSO regression ; logistic regression

Abstract

Detecting variants that predict complex disease is a major goal of human genetics, but is a difficult challenge due to complex underlying etiologies. There are several potential models for the genetic etiology of complex traits, but sources of noise in the data set such as heterogeneity, epistasis, and missing data present challenges for statistical association methods. The effect of these different error-types is largely unknown for data-mining approaches. The impact of each factor needs to be understood to properly apply and modify data mining approaches in human genetics. We simulated data representing a wide range of genetic models to assess the ability of commonly used association analysis approaches to correctly identify the underlying genetic model in the presence of different types of error. Specifically, we compared the power of traditional and stepwise logistic regression, Multifactor Dimensionality Reduction, and LASSO regression and show the relative performance of each method in the presence of these types of noise.

This material is based upon work supported by the National Science Foundation under the NSF-CSUMS project DMS-0703392 (PI: Sujit Ghosh).