The departure between constant-rate birth-death and empirically inferred diversification processes

Birth-death models (BDMs) are stochastic processes describing the processes of speciation and extinction through time and across taxa and are widely used in biology for inference of evolutionary timescales. Previous research has highlighted how the expected trees under BDMs tend to differ from empirical trees with respect to indices such as the amount of phylogenetic imbalance. However, our understanding of how trees differ between BDMs and empirical inferences remains incomplete. In this study, we demonstrate how four different constant-rate BDM scenarios influence tree shape and branch-length characteristics of phylogenetic trees, using a wide range of topology and branch-length indices. Comparison of BDM expectations against a comprehensive empirical dataset of 1,189 empirical trees shows that the dominant form of model inadequacy in BDMs is in failing to accommodate large amounts of phylogenetic imbalance in empirical processes. We also find that empirical trees tend to have significantly greater depth, lower stemminess, and longer shortest pendant edge lengths than BD-simulated trees. The results also indicate that accounting for the sampling fraction is the single most important parameter for accommodating empirical stemminess and branch lengths. Overall, our findings demonstrate the limitations of BDM priors when inferring the shape and structural characteristics of phylogenetic trees, highlighting the importance of novel methods that account for a broader tree space and secondarily reduce any possible bias in branch length estimation.出生-死亡模型（Birth-death models, BDMs）是一类描述随时间推移、跨类群的物种形成与灭绝过程的随机过程，在生物学领域被广泛用于进化时间尺度的推断。既往研究已指出，出生-死亡模型下的期望树在系统发育不平衡性（phylogenetic imbalance）等指标上，往往与经验观测树存在差异。然而，目前学界对出生-死亡模型生成树与经验推断树之间的差异机制仍缺乏完整认知。本研究通过涵盖拓扑结构（topology）与枝长（branch-length）在内的多类指标，阐明了四种不同恒定速率出生-死亡模型场景对系统发育树（phylogenetic trees）形态及枝长特征的影响。将出生-死亡模型的期望树与包含1189棵经验观测树的大规模经验数据集进行对比后发现，出生-死亡模型最突出的不足在于无法适配经验演化过程中存在的高强度系统发育不平衡性。本研究同时发现，经验观测树的树深显著更大、茎占比（stemminess）更低，且最短悬垂枝长度较出生-死亡模型模拟生成的树更长。研究结果还表明，在适配经验树的茎占比与枝长特征时，纳入采样比例参数是最为关键的一环。综上，本研究结果揭示了出生-死亡模型先验在推断系统发育树形态与结构特征时的局限性，同时凸显了两类新方法的重要性：一是能够覆盖更广泛树空间的建模方法，二是可有效降低枝长估计潜在偏差的优化方法。