Significance thresholds for QTL interval mapping tests

The problem of mapping quantitative trait loci (QTL) using genetic marker information is of great interest to the mapping community. There are many statistical methods available for detecting and/or locating QTL, all of which depend on assumptions about the distribution of the quantitative trait values. The distribution of the trait values is affected by sample size, genetic marker density, missing data patterns, environmental noise, etc., all of which affect the distribution of the test statistic used to detect/locate QTL. Failure of the test statistic distribution to follow a standard statistical distribution is the subject of current research. In order to declare a significant QTL effect it is necessary first to understand the behaviour of the test statistic under the null hypothesis so that a critical value may be employed. In this paper we discuss the choices available for obtaining critical values (threshold values) used in locating QTL via interval mapping procedures. We investigate threshold values obtained by different means (analytical approximations and empirical) for the same level of significance (type I error rate) under a normality assumption (null hypothesis of no QTL). In addition, we explore the effect of deviations from normality of the trait values on the threshold value by comparing analytical approximations and empirical threshold values for simulated backcross and F-2 experiments, along with an actual experimental F-2 data set.