Selecting subsets of traits for quantitative trait loci analysis

Identifying genetic determinants of complex traits is a fundamental challenge in genetics research. Historically, a powerful statistical procedure called quantitative trait loci (QTL) mapping has been used to investigate experimental populations for the purpose of finding genomic regions associated with phenotypic traits. When multiple traits are available, there are considerable benefits to analyzing subsets of biologically related traits in a multiple-trait QTL mapping framework. Unfortunately, prior knowledge of which traits are biologically related is often incomplete or missing altogether, which commonly results in each trait being analyzed independently. Single-trait QTL mapping models fail to utilize information from the correlation structure between traits, producing less powerful and less informative hypothesis tests than multiple-trait QTL mapping models, which make it difficult to investigate the relationships between traits. In order to take advantage of the correlation structure between traits, two efficient statistical procedures are proposed to select groups of potentially related traits that can in turn be used in a multiple-trait QTL mapping framework. The first approach identifies groups of traits sharing common sources of variation within a given data set, whereas the second approach identifies groups of traits sharing common sources of variation between fundamentally different categories of data (e.g., phenotypic traits and gene expression abundance). Novel procedures are proposed to address statistical challenges relating to estimating the proper size of these groups and the proper number of groups. Extensive simulation studies are presented to demonstrate the performance of the proposed procedures, and an application is presented to data from the model organism Arabidopsis thaliana