Use of Score Statistics for Model Selection in Mapping Multiple Quantitative Trait Loci

One of the major issues in mapping multiple quantitative trait loci is that of model selection. Many methods for mapping QTL scan positions across the genome and test for statistical significance to decide whether to admit QTL in the model. A common problem with these methods is the determination of a genome-wide threshold value for the test statistic. Recently a modeland data-based resampling procedure using score statistics was introduced to empirically calculate the relevant threshold. In this paper we extend the score statistic analysis to multiple interval mapping (MIM) procedure and perform a simulation study to assess the performance of mapping multiple QTL using a sequential MIM search with model selection based on the use of the score statistic threshold. Our results show that this procedure provides an effective, consistent genome-wide threshold for admitting a QTL to be included in an MIM model. We discuss the performance of the procedure with respect to model size, false discovery rate, power to detect QTL, confidence in position estimates, and behavior of parameter estimates. An important result is that when we relax the genome-wide threshold from the nominal 5% significance level to 20% for model selection, the statistical power to detect QTL can be significantly increased and at the same time the false discovery rate is only mildly affected and is well controlled. Based on the simulation result, we recommend the use of genome-wide threshold at 20% significance level calculated from the score resampling method for model selection in mapping multiple QTL.