We simulate phylogenies and molecular sequences under three different realistic rate variation models-one in which speciation rates and substitution rates both vary but are unlinked, one in which they covary continuously and one punctuated model in which molecular change is concentrated in speciation events, using empirical case studies to parameterise realistic simulations. We test three commonly used \"relaxed clock\" molecular dating methods against these realistic simulations to explore the degree of error in molecular dates under each model. We find average divergence time inference errors ranging from 12% of node age for the unlinked model when reconstructed under an uncorrelated rate prior using BEAST 2, to up to 91% when sequences evolved under the punctuated model are reconstructed under an autocorrelated prior using PAML.
We demonstrate the potential for substantial errors in molecular dates when both speciation rates and substitution rates vary between lineages. This study highlights the need for tests of molecular dating methods against realistic models of rate variation generated from empirical parameters and known relationships.
我们在三种不同的实际速率变化模型下模拟系统发育和分子序列,其中一种是物种形成率和取代率都不同,但没有联系,一种是它们连续地相互变化,一种是分子变化集中在物种形成事件中的标点符号模型,使用经验案例研究来参数化现实模拟。我们针对这些现实的模拟测试了三种常用的“放松时钟”分子测年方法,以探索每种模型下分子日期的误差程度。我们发现,平均发散时间推断误差范围从使用BEAST2在不相关速率下重建时未链接模型的节点年龄的12%到使用PAML在自相关先验下重建在标点模型下进化的序列时高达91%。
我们证明了当物种形成率和取代率在谱系之间变化时,分子日期可能会出现重大错误。这项研究强调了对根据经验参数和已知关系生成的速率变化的现实模型进行分子测年方法测试的必要性。