The Linkage Disequilibrium LASSO for SNP Selection in Genetic Association Studies

• Activity Number: 566
• Type: Topic Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Samuel G. Younkin*+ and J. Sunil Rao and Robert C. Elston and Joseph Nadeau
• Companies: Case Western Reserve University and University of Miami and Case Western Reserve University and Institute for Systems Biology
• Keywords: LASSO ; GWAS ; Alzheimer's ; Linkage Disequilibrium ; R ; SNP

Abstract

A rapid increase in the number of available human genetic variants has led to an agnostic approach to disease-gene mapping which has resulted in the popularization of the genome-wide association study. This method has identified only a small portion of the expected disease susceptibility sites, and it is our belief that many of the remaining SNPs contribute such a small amount to overall heritability that studies cannot overcome problems due to multiple testing. Here we develop a statistical method designed for a genetic association signal that is more representative of the remaining disease susceptibility SNPs. The genetic association signal that we seek is no longer a peak, for the increase in SNP density, coupled with low effect sizes give rise to a plateau-like signal with gaps. We address this by formulating our method as a penalized least squares regression estimator based on the linkage disequilibrium present between SNPs. The method known as the LD LASSO is an adaptation of the fused LASSO used for subset selection when the signal is sparse and block-like. We implement this method in the R package ldlasso, and present results using data from an Alzheimer's GWAS.