Mapping quantitative trait Loci for ordinal traits using the generalized linear model in half-sib designs

Abstract

Most QTL mapping methods share a common assumption: that the phenotype follows a normal distribution. Many phenotypes of interest, however, do not satisfy this assumption. In this paper, a methodology of QTL mapping for ordinal traits based on the framework of the generalized linear model (GLM) is presented. The location and effect of the putative QTL were estimated us- ing the maximum likelihood method. The efficiency and the power of the proposed method were compared with that of the method based on the linear model (LM) in various conditions (QTL ef- fect, heritability, phenotypic incidence, and number of categories of phenotypes) via simulation. A daughter design with multiple families and a total of 500 individuals was applied. The results showed that the GLM approach had certain advantages over the LM approach in power of QTL detection and QTL position estimation for ordinal traits. The estimates of the QTL position were 0.11∼1.59 cM (0.78 on average) less biased with smaller standard errors. The power of QTL detec- tion was 1.6 ∼ 10.9% (5.1% on average) higher. In addition, the power and the accuracy of QTL mapping depended on the effect of the putative quantitative trait loci and the value of heritability. With the increase of the QTL effect from 0.05 to 0.3, the biases of the QTL position estimates reduced 0.4 to 3.6 cM and the power increased 27 to 56% under different heritabilities. With the increase of heritability from 0.1 to 0.4, the biases reduced 0.24 to 3.1 cM and the power increased 5% to 35% under a different QTL effect. ordinal traits / threshold models / generalized linear model / QTL mapping / maximum likelihood