Department of Earth Sciences, Montana State University, Bozeman, MT 59717, USA.
Syst Biol. 2021 Aug 11;70(5):1061-1075. doi: 10.1093/sysbio/syab017.
Phylogenetic comparative methods (PCMs) are commonly used to study evolution and adaptation. However, frequently used PCMs for discrete traits mishandle single evolutionary transitions. They erroneously detect correlated evolution in these situations. For example, hair and mammary glands cannot be said to have evolved in a correlated fashion because each evolved only once in mammals, but a commonly used model (Pagel's Discrete) statistically supports correlated (dependent) evolution. Using simulations, we find that rate parameter estimation, which is central for model selection, is poor in these scenarios due to small effective (evolutionary) sample sizes of independent character state change. Pagel's Discrete model also tends to favor dependent evolution in these scenarios, in part, because it forces evolution through state combinations unobserved in the tip data. This model prohibits simultaneous dual transitions along branches. Models with underlying continuous data distributions (e.g., Threshold and GLMM) are less prone to favor correlated evolution but are still susceptible when evolutionary sample sizes are small. We provide three general recommendations for researchers who encounter these common situations: i) create study designs that evaluate a priori hypotheses and maximize evolutionary sample sizes; ii) assess the suitability of evolutionary models-for discrete traits, we introduce the phylogenetic imbalance ratio; and iii) evaluate evolutionary hypotheses with a consilience of evidence from disparate fields, like biogeography and developmental biology. Consilience plays a central role in hypothesis testing within the historical sciences where experiments are difficult or impossible to conduct, such as many hypotheses about correlated evolution. These recommendations are useful for investigations that employ any type of PCM. [Class imbalance; consilience; correlated evolution; evolutionary sample size; phylogenetic comparative methods.].
系统发育比较方法（PCM）常用于研究进化和适应。然而，离散特征常用的 PCM 处理不了单一进化转变。在这些情况下，它们错误地检测到相关进化。例如，毛发和乳腺不能被说成是以相关的方式进化的，因为它们在哺乳动物中只进化了一次，但是一个常用的模型（Pagel 的离散）在统计上支持相关（依赖）进化。通过模拟，我们发现，对于模型选择至关重要的速率参数估计在这些情况下很差，因为独立特征状态变化的有效（进化）样本量很小。Pagel 的离散模型在这些情况下也倾向于支持依赖进化，部分原因是它迫使进化通过在尖端数据中未观察到的状态组合进行。这个模型禁止分支上同时发生双重转变。具有潜在连续数据分布的模型（例如阈值和 GLMM）不太容易偏向相关进化，但当进化样本量较小时仍然容易受到影响。我们为遇到这些常见情况的研究人员提供了三个一般建议：i）创建研究设计，评估先验假设并最大限度地增加进化样本量；ii）评估进化模型的适用性-对于离散特征，我们引入了系统发育不平衡比；iii）通过来自不同领域的证据的一致性，如生物地理学和发育生物学，评估进化假设。在历史科学中，证据的一致性在假设检验中起着核心作用，因为在这些科学中，实验是困难或不可能进行的，例如许多关于相关进化的假设。这些建议对于使用任何类型的 PCM 的研究都是有用的。[类不平衡；一致性；相关进化；进化样本量；系统发育比较方法。]
