Relative Efficacy of GWAS Models for Ordinal Variables

Genome wide association mapping (GWAS) represents one of the most effective ways to detect quantitative trait loci (QTL). Conventional GWAS models assume that phenotypic variables fall on a continuous scale. When those models are applied to ordinal data, for example, GWAS results might be compromised by out of range predictions, heterogeneity of variances and inaccurate hypothesis testing results. There are several alternatives for mapping QTL of ordinal data through GWAS that include data transformation, generalized linear model, and non-parametric models). However, the empirical benefit of one method over another is not entirely clear. To answer this, five methods for GWAS for ordinal variables were compared in terms of their relative performance to detect QTL using simulated and real data. On a wide range of population sizes, number of QTL and heritabilities, no differences for power and false discovery rate were detected across methods, while similar bias in the estimation of QTL effects was obtained for all methods. This suggests that a classic GWAS model for ordinal data could still be appropriate for QTL mapping without compromising the efficacy of QTL detection.