Bray-Curtis (AFD) differentiation in molecular ecology: Forecasting, an adjustment (AA), and comparative performance in selection detection

Abstract

Geographic genetic differentiation measures are used for purposes such as assessing genetic diversity and connectivity, and searching for signals of selection. Confirmation by unrelated measures can minimize false positives. A popular differentiation measure, Bray-Curtis, has been used increasingly in molecular ecology, renamed AFD (hereafter called BCAFD). Critically, BCAFD is expected to be partially independent of the commonly used Hill “Q-profile” measures. BCAFD needs scrutiny for potential biases, by examining limits on its value, and comparing simulations against expectations. BCAFD has two dependencies on within-population (alpha) variation, undesirable for a between-population (beta) measure. The first dependency is derived from similarity to (Formula presented.) and (Formula presented.). The second dependency is that BCAFD cannot be larger than the highest allele proportion in either location (alpha variation), which can be overcome by data-filtering or by a modified statistic AA or “Adjusted AFD”. The first dependency does not forestall applications such as assessing connectivity or selection, if we know the measure's null behavior under selective neutrality with specified conditions—which is shown in this article for AA, for equilibrium, and nonequilibrium, for the commonly used data type of single-nucleotide-polymorphisms (SNPs) in two locations. Thus, AA can be used in tandem with mathematically contrasting differentiation measures, with the aim of reducing false inferences. For detecting adaptive loci, the relative performance of AA and other measures was evaluated, showing that it is best to use two mathematically different measures simultaneously, and that AA is in one of the best such pairwise criteria. For any application, using AA, rather than BCAFD, avoids the counterintuitive limitation by maximum allele proportion within localities.