A simulation study on the performance of differentiation-based methods to detect selected loci using linked neutral markers

We investigated the performance of two of the most popular differentiation-based methods to detect loci under selection (dfdist/fdist and bayescan) in order to ascertain the average chromosome map distance between the detected outlier markers and the nearest loci under selection. We used a model of neutral markers genetically linked to selected loci (QTL) controlling a quantitative trait subject to divergent selection in two subpopulations connected by migration. The results are not particularly encouraging because for chromosome lengths above 0.5 morgan, at least 30% of outliers detected were positioned in chromosomes where QTL were absent, clearly denoting false positives. Outliers linked to QTL were on average closer to the nearest QTL than randomly chosen markers, but the methods showed a substantial uncertainty about the genetic association between markers and selected loci, as this association could be shown significantly only in a moderate number of replicates for most scenarios. At equal conditions, bayescan seemed to perform somewhat more efficiently than dfdist/fdist, with little difference between results for dominant and codominant markers.