A new Bayesian automatic model selection approach for mapping quantitative trait loci under variance component model

Bayesian variable selection implemented via reversible jump Markov chain Monte Carlo (RJMCMC) is an effective method for mapping multiple QTL. Recently, it has been used for QTL mapping both in inbred line crosses and in outbred populations. However, by RJMCMC, since the model-dimension is variable, the parameters are usually subject to poor mixing and difficult to converge. In inbred lines, the fixed effect model is used for mapping QTL, various approaches which keep the model-dimension unchanged have been proposed, and it is proved that the mixing properties of Markov chains is substantially improved compared with RJMCMC. In outbred populations, the random effect model is used and the implementation via RJMCMC for variable selection still is the mainstream to map multiple QTL. Due to the poor performance RJMCMC has, it is meaningful to develop a model-dimension fixed approach for mapping QTL under random effect model. In this article, we proposed a new model-dimension fixed approach called Bayesian automatic model selection method for mapping multiple QTL under random effect model. By the new approach, all variances of QTL are subject to estimate, in which the variance of zero-effect QTL will exactly converge to zero, and those of non-zero effect QTL will be estimated precisely. Therefore, no special model selection is required. A series of simulation experiments have been conducted to investigate the performance of the method, the result showed that the new approach is very efficient for mapping multiple QTL. A computer program written in FORTRAN is available to interested users on request.