Abstract:
|
Detecting natural selection in temporal genetic samples has many uses, from revealing the events that accompany cross-species transmissions, the changes that presage effective immune defenses, how modern species evolved from their ancient ancestors, to how selective breeding programs work. Estimation of selection is complicated by the need to account for nuisance parameters, such as effective population size or the time lapse. Nishino's (2013) FITR is a simple test for detecting selection that approximates relative allele frequency changes in time using a Gaussian distribution. It accounts for nuisance parameters by utilizing data known to come from unselected, neutral alleles. We extend the method by using normalizing transformations, the Box-Cox transformation proving to be the best. Special accommodation is made for allele fixation, which associates with non-normality and violation of FITR assumptions. We show the proposed method achieves type I error near the nominal level with considerably more power than the original FITR method, producing far superior ROC curves for predicting selected sites. We apply the method to several datasets and interpret our findings.
|