Designs and solutions to multiple trait comparisons

Abstract

Multiple trait detection and analysis of quantitative trait loci via linkage to genetic markers is problematic, first because of the increase in the number of comparisons tested, second because of possible multitrait loci effects, and third because of biases due to selection based on phenotypic trait values. Nearly all studies that considered multiple traits have analyzed each trait separately. Two methods have been proposed for multitrait experiments; maximum likelihood multivariate analysis, and canonical transformation to a set of uncorrelated variables. If individuals are selected for genotyping based on a single trait, parameter estimates for other correlated traits will be biased using single trait analyses, and significance levels will be incorrect. With multivariate analysis, unbiased estimates of QTL effects can be derived even with selective genotyping. Furthermore, power is increased per individual genotyped, even if selective genotyping is relative to a trait unaffected by the segregating locus. For a preliminary multiple trait analysis, controlling the false discovery rate rather than the experiment wise type‐I error allows for greater statistical power to detect true effects. In the absence of any true effects the two methods are equivalent. An example is given using granddaughter design data.