A unified semiparametric framework for quantitative trait loci analyses, with application to spike phenotypes

This article proposes a general semiparametric model for multiple quantitative trait loci (QTL) analyses of complex phenotypes in backcross and intercross designs. The model provides tests about genetic hypotheses, such as additivity, dominance, and epistasis, that do not require specifying the form of the phenotypic distribution. This contrasts with previous approaches based on transformations to normality and generalized linear models, which require careful consideration of the phenotypic distribution. Inferences involve extensions of partial and conditional likelihoods developed for single-QTL backcross models. We demonstrate that conditional likelihood is robust to unobserved selective genotyping, whereas partial likelihood and other standard methods are not. To facilitate genome screens, a novel resampling method is proposed that is similar in spirit to the popular permutation tests. Its main advantages are that it is broadly applicable to multiple QTLs with nonnormal phenotypes and achieves a substantial reduction in computational burden. A thorough case study of spike data on the genetic influences to recovery from Listeria infection in a mouse intercross experiment is presented. The application reveals that the proposed methods may give substantively different conclusions than those obtained with existing interval mapping methods from parametric models in the presence of unobserved selection.