Cladistic permutation tests for monophyly and nonmonophyly
Cladistic permutation tests for monophyly and nonmonophyly
复制标题
DOI：
10.1093/sysbio/40.3.366
复制
发表时间：
1991-09
影响因子：
6.5
通讯作者：
D. Faith
中科院分区：
文献类型：
--
作者：
D. Faith
最近关于分支结构的随机化或排列测试的工作(Archie，1989a，1989b；Faith，1990；Faith and Cranston，1991)揭示了一些戏剧性的案例，在这些案例中，表面上具有系统发育信息的数据实际上具有仅凭机会就可以很容易匹配的结构。在这些测试中，使用了特定的空模型，其中每个字符的状态被随机地重新分配给分类群，从而所产生的随机化数据集表示字符之间的随机协变。相应的零假设是，可以为这种随机化数据找到观察到的分层结构。随机性-协变零模型的一个必要的伴随者是用于评估和量化等级结构的某种标准。这种对结构的测量代表了一个标准模型，该模型规定了数据预期与潜在模式相关的方式(Faith和Cranston，1991)。在上面提到的例子中，等级结构是通过相应的最小长度树的长度来测量的(在该测试的一般版本中已经使用了等级结构的其他测量；见Faith[1991])。分支学简约标准的选择意味着，零假设是通过比较观察到的最小长度与许多随机数据集的最小长度来评估的。所有集合(观察到的和随机的)的长度与观察到的长度一样短或短于观察到的长度的比例产生了“分支排列尾概率”，或PTP(Faith和Cranston，1991；相同的测试是由Archie提出的独立Lyl[1989a])。例如，如果5%或更少的数据集的长度等于或小于原始数据集的长度(PTP<；0.05)，则可以在通常的0.05水平上拒绝零假设。这一测试的最初应用之一(Faith，1990)和对它的答复(Thomas等人，1990)提出了一些重要的争议点，本文将予以论述。在早期的一篇论文中，Thomas等人。(1989)使用12S核糖体RNA基因序列数据来探索南美和澳大利亚有袋类之间的关系，解决了关于假定灭绝的袋猴(Thylacinus)与这些其他类群的关系的长期争议。基于对这些数据的分支分析，他们认为袋狼不是澳大利亚有袋类动物的外类群，而是Dasyuridae的姐妹。然而，PTP检验的应用表明，该数据集只包含7个系统发育信息特征(表1a)，没有显著的分支结构。因此，12S核糖体RNA基因序列数据不足以推断袋狼与其他分类群的系统发育关系(Faith，1990)。在他们的答复中，Thomas等人。(1990)认为，尽管整个数据集没有明显的等级结构，但这与他们感兴趣的假设没有直接关系，在他们的答复中，这是关于袋狼和蝙蝠类单系的具体问题。他们声称，对于这种单调性假设的评估，Bootstrap(Felsenstein，1985)是一个更强大的检验(另见Archie，1989a)。Bootstrap检验的应用表明，他们的假设得到了显著支持
Recent work on randomization or permutation tests for cladistic structure (Archie, 1989a, 1989b; Faith, 1990; Faith and Cranston, 1991) has revealed some dramatic cases in which apparently phylogenetically informative data in fact have structure that could easily be matched by chance alone. In these tests a particular null model is used in which each character's states are reassigned randomly to the taxa, so that the resulting randomized data set represents random covariation among the characters. The corresponding null hypothesis is that the observed hierarchical structure could be found for such randomized data. A necessary companion to the randomcharacter-covariation null model is some criterion for evaluating and quantifying hierarchical structure. This measure of structure represents a criterion model that prescribes the manner in which the data are expected to relate to the underlying pattern (Faith and Cranston, 1991). In the examples referred to above, hierarchical structure is measured by the length of the corresponding minimum-length tree (other measures of hierarchical structure have been used in a general version of this test; see Faith [1991]). The choice of the parsimony criterion of cladistics implies that the null hypothesis is evaluated by comparing the observed minimum length to that achieved for many randomized data sets. The proportion of all sets (observed and random) having lengths as short as or shorter than the observed length yields the "cladistic permutation tail probability," or PTP (Faith and Cranston, 1991; the same test was independentlyl proposed by Archie [1989a]). The null hypothesis may be rejected, for example at the usual 0.05 level, if 5% or fewer of the data sets have a length equal to or less than that of the original (PTP < 0.05). One of the initial applications of this test (Faith, 1990) and the reply to it (Thomas et al., 1990) have raised some important points of controversy that will be addressed in this paper. In an earlier paper, Thomas et al. (1989) used 12S ribosomal RNA gene sequence data to explore the relationships among South American and Australian marsupials, addressing longstanding controversies about the relationship of the presumed-extinct thylacine (Thylacinus) to these other taxa. Based on their cladistic analysis of these data, they argued that the thylacine is not an outgroup to the Australian marsupials, and is the sister of the Dasyuridae. However, application of the PTP test showed that this data set, which consisted of only seven phylogenetically informative characters (Table la), did not have significant cladistic structure. Thus, the 12S ribosomal RNA gene sequence data constituted insufficient evidence for the inference of the phylogenetic relationship of the thylacine to these other taxa (Faith, 1990). In their reply, Thomas et al. (1990) argued that although the data set as a whole did not have significant hierarchical structure, this was not directly relevant to their hypothesis of interest, which was represented in their reply as the specific question of the monophyly of thylacines and dasyurids. They claimed that, for the evaluation of such hypotheses of monophyly, the bootstrap (Felsenstein, 1985) was a more powerful test (see also Archie, 1989a). Application of the bootstrap test revealed significant support for their hypothesis of