Department of Ecology and Evolutionary Biology, University of Michigan, 830 N. University, Ann Arbor, MI 48109, USA.
Syst Biol. 2018 Mar 1;67(2):328-339. doi: 10.1093/sysbio/syx072.
The recent surge in enthusiasm for simultaneously inferring relationships from extinct and extant species has reinvigorated interest in statistical approaches for modeling morphological evolution. Current statistical methods use the Mk model to describe substitutions between discrete character states. Although representing a significant step forward, the Mk model presents challenges in biological interpretation, and its adequacy in modeling morphological evolution has not been well explored. Another major hurdle in morphological phylogenetics concerns the process of character coding of discrete characters. The often subjective nature of discrete character coding can generate discordant results that are rooted in individual researchers' subjective interpretations. Employing continuous measurements to infer phylogenies may alleviate some of these issues. Although not widely used in the inference of topology, models describing the evolution of continuous characters have been well examined, and their statistical behavior is well understood. Also, continuous measurements avoid the substantial ambiguity often associated with the assignment of discrete characters to states. I present a set of simulations to determine whether use of continuous characters is a feasible alternative or supplement to discrete characters for inferring phylogeny. I compare relative reconstruction accuracy by inferring phylogenies from simulated continuous and discrete characters. These tests demonstrate significant promise for continuous traits by demonstrating their higher overall accuracy as compared to reconstruction from discrete characters under Mk when simulated under unbounded Brownian motion, and equal performance when simulated under an Ornstein-Uhlenbeck model. Continuous characters also perform reasonably well in the presence of covariance between sites. I argue that inferring phylogenies directly from continuous traits may be benefit efforts to maximize phylogenetic information in morphological data sets by preserving larger variation in state space compared to many discretization schemes. I also suggest that the use of continuous trait models in phylogenetic reconstruction may alleviate potential concerns of discrete character model adequacy, while identifying areas that require further study in this area. This study provides an initial controlled demonstration of the efficacy of continuous characters in phylogenetic inference.
近期，人们对同时推断灭绝物种和现存物种之间关系的兴趣大增，这重新激发了人们对用于模拟形态进化的统计方法的兴趣。当前的统计方法使用 Mk 模型来描述离散特征状态之间的替换。尽管这代表了向前迈出的重要一步，但 Mk 模型在生物学解释方面存在挑战，其在模拟形态进化方面的充分性尚未得到很好的探索。形态系统发生学的另一个主要障碍涉及离散特征编码的特征编码过程。离散特征编码的主观性往往会产生源于个别研究人员主观解释的不一致结果。使用连续测量值推断系统发育可能会缓解其中的一些问题。尽管在拓扑推断中未广泛使用，但描述连续特征进化的模型已经得到了很好的检验，并且其统计行为也得到了很好的理解。此外，连续测量值避免了与将离散特征分配给状态相关的大量歧义。我提出了一组模拟实验，以确定使用连续特征是否是推断系统发育的离散特征的可行替代或补充。我通过从模拟的连续和离散特征推断系统发育来比较相对重建准确性。这些测试通过在模拟无界布朗运动下，与 Mk 相比，显示出连续特征的整体准确性更高，而在模拟奥恩斯坦 - 乌伦贝克模型下，性能相等，从而为连续特征提供了很大的希望。在存在站点之间协方差的情况下，连续特征的表现也相当不错。我认为，与离散特征相比，直接从连续特征推断系统发育可能会通过与许多离散化方案相比，保留更大的状态空间变化，从而有助于最大限度地提高形态数据集的系统发育信息。我还建议，在系统发育重建中使用连续特征模型可能会减轻离散特征模型充分性的潜在问题，同时确定该领域需要进一步研究的领域。本研究提供了连续特征在系统发育推断中的功效的初步对照演示。
