Parametric and semiparametric methods for mapping quantitative trait loci

Most statistical methods for mapping quantitative trait loci (QTLs) have been extensively developed for normally distributed and completely observed phenotypes. A survival model such as Cox's proportional hazards model, or an accelerated failure time model, is the natural choice for regressing a phenotype onto a maker-type when the primary phenotype belongs to failure time. This study proposes parametric and semiparametric methods based on accelerated failure time models for interval mapping. In the parametric model, the EM algorithm (expectation maximization algorithm) is adopted to estimate regression parameters and search the entire chromosome for QTL. In the semiparametric model, the error distribution is left unspecified, and a rank-based inference is developed for the effect and location of QTL. Simulated data are used to compare the mapping performance of the semiparametric methodology with that of the parametric method, which separately selects the correct and incorrect error distribution. Analytical results reveal that the parametric estimators may be more efficient in determining the effect and location of QTL, but have obvious bias when selecting the incorrect error distribution. In contrast, the semiparametric inference is robust to the error distribution.