International Biometric Society Partial-Least-Squares-Regression models for genome-wide association studies

The power of genome-wide association (GWA) studies to detect associations between genetic variants and disease susceptibility has been amply demonstrated in the recent series of papers detailing the discovery of significantly-associated variants that have successfully replicated (e.g. [1, 2]). Recent technological advancements in genotyping have led to drastic decreases in the cost of collecting individual genetic variation data, in particular Single-Nucleotide-Polymorphism (SNP) data. As a result, typical GWA studies currently include the genetic information from hundreds of thousands of SNPs collected in thousands of individuals, with these numbers likely to increase even further over the next few years. Such studies bring new challenges in statistical analysis, including the need to develop methods that can deal efficiently with the vast amounts of data and account for the complex correlation structure amongst SNPs across the genome. Dealing with the potentially high correlations amongst SNPs in densely genotyped regions is likely to become even more of a prominent factor as studies shift from collecting less correlated “tag-SNP ” data (where SNPs are often selected to have, e.g., pairwise correlation coeffients below some threshold) to exhaustive “resequencing ” data (where all SNPs amongst the cohort individuals are collected)