Robustness of the maximum-likelihood-binomial approach for linkage analysis of quantitative trait loci with non-normal phenotypic data

Abstract

Introduction Model-free linkage studies are increasingly used to investigate the genetic factors implicated in complex quantitative traits because they do not require any specification of the underlying genetic model. However, the term model-free does not imply that no assumption is introduced by the corresponding statistical methods. In particular, the widely used variance components approaches assume multivariate normality of the phenotypic distribution and it has been shown that violation of this normality hypothesis could lead to large inflation of the type I error. In this paper, we assess the robustness of the recently developed sibship-oriented Maximum-Likelihood-Binomial (MLB) method for genetic model-free linkage analysis in the context of several types of non-normal phenotypic data using a large simulation study.

Simulation study Under the hypothesis of no linkage at the marker locus under study, 20 000 replicates of family samples including 100 or 500 independent sib-pairs were simulated considering four different designs that lead to non-normal phenotypic data: (1) presence of a major gene not linked to the studied marker, (2) gene–environment interaction, (3) analysis of a binary phenotype, and (4) extreme sampling. Further, three levels of residual sib–sib correlation were considered.

Results and discussion For each simulation design the empirical type I errors were consistent with their asymptotic expectations showing that the MLB approach is insensitive to non-normal phenotypic distribution whatever the mechanism underlying this non-normality. Therefore, the MLB method should be an attractive alternative method for model-free linkage analysis of QTL, especially for investigators who do not want to worry about the validity of asymptotic thresholds when performing their analyses.