Ronquist Fredrik, Lartillot Nicolas, Phillips Matthew J
Department of Bioinformatics and Genetics, Swedish Museum of Natural History, PO Box 50007, 104 05 Stockholm, Sweden
Laboratoire de Biométrie et Biologie Evolutive, UMR CNRS 5558, Université Claude Bernard Lyon 1, F-69622 Villeurbanne Cedex, France.
Philos Trans R Soc Lond B Biol Sci. 2016 Jul 19;371(1699). doi: 10.1098/rstb.2015.0136.
Total-evidence dating (TED) allows evolutionary biologists to incorporate a wide range of dating information into a unified statistical analysis. One might expect this to improve the agreement between rocks and clocks but this is not necessarily the case. We explore the reasons for such discordance using a mammalian dataset with rich molecular, morphological and fossil information. There is strong conflict in this dataset between morphology and molecules under standard stochastic models. This causes TED to push divergence events back in time when using inadequate models or vague priors, a phenomenon we term 'deep root attraction' (DRA). We identify several causes of DRA. Failure to account for diversified sampling results in dramatic DRA, but this can be addressed using existing techniques. Inadequate morphological models also appear to be a major contributor to DRA. The major reason seems to be that current models do not account for dependencies among morphological characters, causing distorted topology and branch length estimates. This is particularly problematic for huge morphological datasets, which may contain large numbers of correlated characters. Finally, diversification and fossil sampling priors that do not incorporate all the available background information can contribute to DRA, but these priors can also be used to compensate for DRA. Specifically, we show that DRA in the mammalian dataset can be addressed by introducing a modest extra penalty for ghost lineages that are unobserved in the fossil record, for instance by assuming rapid diversification, rare extinction or high fossil sampling rate; any of these assumptions produces highly congruent divergence time estimates with a minimal gap between rocks and clocks. Under these conditions, fossils have a stabilizing influence on divergence time estimates and significantly increase the precision of those estimates, which are generally close to the dates suggested by palaeontologists.This article is part of the themed issue 'Dating species divergences using rocks and clocks'.
全证据定年法（TED）使进化生物学家能够将广泛的定年信息纳入统一的统计分析中。人们可能会认为这会提高岩石年代与分子钟之间的一致性，但情况未必如此。我们使用一个包含丰富分子、形态和化石信息的哺乳动物数据集来探究这种不一致的原因。在标准随机模型下，该数据集中形态学和分子数据之间存在强烈冲突。这导致在使用不充分的模型或模糊的先验条件时，TED会将分歧事件的时间推回到过去，我们将这种现象称为“深根吸引”（DRA）。我们确定了DRA的几个原因。未能考虑多样化采样会导致显著的DRA，但可以使用现有技术来解决这一问题。不充分的形态学模型似乎也是DRA的一个主要因素。主要原因似乎是当前模型没有考虑形态特征之间的相关性，从而导致拓扑结构和分支长度估计出现偏差。对于庞大的形态学数据集来说，这一问题尤为突出，因为其中可能包含大量相关特征。最后，未纳入所有可用背景信息的多样化和化石采样先验条件可能会导致DRA，但这些先验条件也可用于补偿DRA。具体而言，我们表明，通过对化石记录中未观察到的幽灵谱系引入适度的额外惩罚，可以解决哺乳动物数据集中的DRA问题，例如假设快速多样化、罕见灭绝或高化石采样率；这些假设中的任何一个都能产生高度一致的分歧时间估计，且岩石年代与分子钟之间的差距最小。在这些条件下，化石对分歧时间估计具有稳定作用，并显著提高了这些估计的精度，这些估计通常接近古生物学家提出的日期。本文是主题为“利用岩石和时钟确定物种分歧时间”的特刊的一部分。
