Goolsby Eric W
Interdisciplinary Toxicology Program, Department of Plant Biology, University of Georgia, Athens GA, 30602, USA
Syst Biol. 2015 Jul;64(4):568-78. doi: 10.1093/sysbio/syv012. Epub 2015 Feb 10.
Phylogenetic comparative methods offer a suite of tools for studying trait evolution. However, most models inherently assume fixed trait values within species. Although some methods can incorporate error around species means, few are capable of accounting for variation driven by environmental or temporal gradients, such as trait responses to abiotic stress or ontogenetic trajectories. Such traits, often referred to as function-valued or infinite-dimensional, are typically expressed as reaction norms, dose-response curves, or time plots and are described by mathematical functions linking independent predictor variables to the trait of interest. Here, I introduce a method for extending ancestral state reconstruction to incorporate function-valued traits in a phylogenetic generalized least squares (PGLS) framework, as well as extensions of this method for testing phylogenetic signal, performing phylogenetic analysis of variance (ANOVA), and testing for correlated trait evolution using recently proposed multivariate PGLS methods. Statistical power of function-valued comparative methods is compared to univariate approaches using data simulations, and the assumptions and challenges of each are discussed in detail.
系统发育比较方法提供了一套用于研究性状进化的工具。然而，大多数模型本质上假设物种内的性状值是固定的。虽然一些方法可以纳入围绕物种均值的误差，但很少有方法能够解释由环境或时间梯度驱动的变异，例如性状对非生物胁迫的响应或个体发育轨迹。这类性状通常被称为函数值性状或无限维性状，通常表示为反应规范、剂量反应曲线或时间图，并通过将独立预测变量与感兴趣的性状联系起来的数学函数来描述。在这里，我介绍一种方法，用于扩展祖先状态重建，以便在系统发育广义最小二乘法（PGLS）框架中纳入函数值性状，以及该方法的扩展，用于测试系统发育信号、进行系统发育方差分析（ANOVA），以及使用最近提出的多变量PGLS方法测试相关性状进化。使用数据模拟将函数值比较方法的统计功效与单变量方法进行比较，并详细讨论了每种方法的假设和挑战。
