Identifying Signatures of Selection in Genetic Time Series

摘要

Both genetic drift and natural selection cause the frequencies of alleles ina population to vary over time. Discriminating between these two evolutionaryforces, based on a time series of samples from a population, remains anoutstanding problem with increasing relevance to modern data sets. Even in theidealized situation when the sampled locus is independent of all other locithis problem is difficult to solve, especially when the size of the populationfrom which the samples are drawn is unknown. A standard $\chi^2$-basedlikelihood ratio test was previously proposed to address this problem. Here weshow that the $\chi^2$ test of selection substantially underestimates theprobability of Type I error, leading to more false positives than indicated byits $P$-value, especially at stringent $P$-values. We introduce two methods tocorrect this bias. The empirical likelihood ratio test (ELRT) rejectsneutrality when the likelihood ratio statistic falls in the tail of theempirical distribution obtained under the most likely neutral population size.The frequency increment test (FIT) rejects neutrality if the distribution ofnormalized allele frequency increments exhibits a mean that deviatessignificantly from zero. We characterize the statistical power of these twotests for selection, and we apply them to three experimental data sets. Wedemonstrate that both ELRT and FIT have power to detect selection in practicalparameter regimes, such as those encountered in microbial evolutionexperiments. Our analysis applies to a single diallelic locus, assumedindependent of all other loci, which is most relevant to full-genome selectionscans in sexual organisms, and also to evolution experiments in asexualorganisms as long as clonal interference is weak. Different techniques will berequired to detect selection in time series of co-segregating linked loci.