Tip rates, phylogenies, and diversification: what are we estimating, and how good are the estimates?

1. Species-specific diversification rates, or “tip rates”, can be computed quickly from phylogenies and are widely used to study diversification rate variation in relation to geography, ecology, and phenotypes. These tip rates provide a number of theoretical and practical advantages, such as the relaxation of assumptions of rate homogeneity in trait-dependent diversification studies. However, there is substantial confusion in the literature regarding whether these metrics estimate speciation or net diversification rates. Additionally, no study has yet compared the relative performance and accuracy of tip rate metrics across simulated diversification scenarios.
2. We compared the statistical performance of three model-free rate metrics (inverse terminal branch lengths; node density metric; DR statistic) and a model-based approach (BAMM). We applied each method to a large set of simulated phylogenies that had been generated under different diversification processes. We summarized performance in relation to the type of rate variation, the magnitude of rate heterogeneity and rate regime size. We also compared the ability of the metrics to estimate both speciation and net diversification rates.
3. We show decisively that model-free tip rate metrics provide a better estimate of the rate of speciation than of net diversification. Error in net diversification rate estimates increases as a function of the relative extinction rate. In contrast, error in speciation rate estimates is low and relatively insensitive to extinction. Overall, and in particular when relative extinction was high, BAMM inferred the most accurate tip rates and exhibited lower error than non-model-based approaches. DR was highly correlated with true speciation rates but exhibited high error variance, and was the best metric for very small rate regimes.
4. We found that, of the metrics tested, DR and BAMM are the most useful metrics for studying speciation rate dynamics and trait-dependent diversification. Although BAMM was more accurate than DR overall, the two approaches have complementary strengths. Because tip rate metrics are more reliable estimators of speciation rate, we recommend that empirical studies using these metrics exercise caution when drawing biological interpretations in any situation where the distinction between speciation and net diversification is important.

1. 物种特异性分化速率，或称“末端速率（tip rates）”，可快速从系统发育树（phylogenies）中计算得到，被广泛用于探究与地理、生态及表型相关的分化速率变异。这类末端速率具备多项理论与实践优势，例如可放宽性状依赖型分化研究中速率均一性的假设前提。但当前学界相关文献中，对这类指标究竟是估算物种形成速率还是净分化速率（net diversification rates）仍存在大量认知混淆。此外，尚无研究在模拟分化场景下对各类末端速率指标的相对性能与准确度展开对比。
2. 本研究对比了三类无模型速率指标（末端分支长度倒数法、节点密度指标、DR统计量（DR statistic））与一种基于模型的方法（BAMM）的统计性能。我们将每种方法应用于大量基于不同分化过程生成的模拟系统发育树数据集，并针对速率变异类型、速率异质性程度及速率组规模，对各方法的性能进行了总结。此外，我们还对比了各类指标对物种形成速率与净分化速率的估算能力。
3. 本研究明确证实，无模型末端速率指标对物种形成速率的估算效果优于净分化速率。净分化速率估算误差随相对灭绝速率的升高而增大。与之相反，物种形成速率的估算误差较低，且相对不受灭绝事件的影响。总体而言，尤其是在相对灭绝速率较高的场景下，BAMM能够得到最准确的末端速率估算结果，且误差低于各类无模型方法。DR统计量与真实物种形成速率具有极高的相关性，但误差方差较大，且是针对极小速率组的最优指标。
4. 本研究发现，在所测试的指标中，DR统计量与BAMM是探究物种形成速率动态及性状依赖型分化的最实用指标。尽管整体而言BAMM的准确度优于DR统计量，但二者的优势具有互补性。鉴于末端速率指标对物种形成速率的估算更为可靠，我们建议使用这类指标的实证研究在涉及物种形成速率与净分化速率区分的关键场景中，进行生物学解读时需格外谨慎。