BayesQTLBIC: Bayesian multi-locus QTL analysis based on the BIC criterion

R package for a non-MCMC approximate multilocus Bayesian model selection approach to the analysis of quantitative trait loci (QTL). The method and models are described in (Ball, R. D. Genetics 159: 1351–1364, 2001; http://www.genetics.org/cgi/content/abstract/159/3/1351). Data is assumed to be from a QTL mapping family with DNA markers genotyped along the genome. The QTL mapping problem is represented as a model selection problem, where each model is a linear regression of the trait on a selected set of marker values. The main function bicreg.qtl() is based on the S function bicreg()— posterior probabilities for models are approximated from the BIC criterion, calculated for each model in a search of model space using leaps or regsubsets. Additionally, we allow for prior probabilities based on expected numbers of QTL per genome and options to control the size of models considered, and to allow for selectivly genotyping from the tails of the phenotypic distribution. Missing values are estimated by multiple imputation, and estimates of marker effects can be obtained conditional on selection or unconditional and free of selection bias. The method relies on 3 approximations: (1.) QTL configuration is represented approximately by configurations with QTL located at marker positions; (2.) Posterior probabilities are given approximately in terms of the BIC criterion; and (3.) The distribution of missing marker values is approximated by multiple imputation, sampling from the distribution of missing values conditional on non-missing values. We have found these are good approximations provided (1.) the marker spacing is reasonable (less than 30cM); (2.) the sample size is 100 or more for fully genotyped populations; and (3.) around 10 imputations are used and the effect of any given QTL on the trait is not large. Due to limits on the number of markers that can be considered simultaneously the method is generally applied separately to each chromosome or could be iteratively applied to sets of chromosomes using fixed sets of predictors from other chromsomes when analysing a given chromosome.

Downloads: