Centre National de la Recherche Scientifique–Muséum National d'Histoire Naturelle, Brunoy, France.
Genetics. 2011 May;188(1):165-79. doi: 10.1534/genetics.110.121764. Epub 2011 Mar 8.
Reconstructing the demographic history of populations is a central issue in evolutionary biology. Using likelihood-based methods coupled with Monte Carlo simulations, it is now possible to reconstruct past changes in population size from genetic data. Using simulated data sets under various demographic scenarios, we evaluate the statistical performance of Msvar, a full-likelihood Bayesian method that infers past demographic change from microsatellite data. Our simulation tests show that Msvar is very efficient at detecting population declines and expansions, provided the event is neither too weak nor too recent. We further show that Msvar outperforms two moment-based methods (the M-ratio test and Bottleneck) for detecting population size changes, whatever the time and the severity of the event. The same trend emerges from a compilation of empirical studies. The latest version of Msvar provides estimates of the current and the ancestral population size and the time since the population started changing in size. We show that, in the absence of prior knowledge, Msvar provides little information on the mutation rate, which results in biased estimates and/or wide credibility intervals for each of the demographic parameters. However, scaling the population size parameters with the mutation rate and scaling the time with current population size, as coalescent theory requires, significantly improves the quality of the estimates for contraction but not for expansion scenarios. Finally, our results suggest that Msvar is robust to moderate departures from a strict stepwise mutation model.
重建人口的历史是进化生物学中的一个核心问题。现在，通过基于似然的方法结合蒙特卡罗模拟，可以从遗传数据中重建过去的种群大小变化。我们使用各种人口统计场景下的模拟数据集，评估了 Msvar 的统计性能，这是一种从微卫星数据推断过去人口变化的全似然贝叶斯方法。我们的模拟测试表明，只要事件既不太弱也不太近，Msvar 就能非常有效地检测到种群的减少和扩张。我们进一步表明，无论事件的时间和严重程度如何，Msvar 都优于两种基于矩的方法（M-ratio 检验和瓶颈）来检测种群大小的变化。从一组经验研究的汇编中也出现了同样的趋势。Msvar 的最新版本提供了当前和祖先种群大小以及种群开始改变大小的时间的估计。我们表明，在没有先验知识的情况下，Msvar 对突变率提供的信息很少，这导致每个人口统计参数的估计值存在偏差和/或置信区间较宽。然而，按照合并理论的要求，用突变率缩放种群大小参数，并将时间与当前种群大小缩放，显著提高了收缩场景下估计值的质量，但不能提高扩张场景下的质量。最后，我们的结果表明，Msvar 对严格逐步突变模型的适度偏离具有鲁棒性。
