分子系统进化关系分析的一种新方法--贝叶斯法在硬蜱属中的应用

距离矩阵邻接法、最大简约法和最大似然法是重建生物系统关系的3种主要方法。普遍认为最大似然法在原理上优于前二种方法，但其计算复杂费时。由于现行计算机的能力尚达不到其要求而实用性差，特别是在处理大数据集样本(即大于25个分类单元)时，用此方法几乎不可能。新近提出的贝叶斯法(Bayesianmethod)既保留了最大似然法的基本原理，又引进了马尔科夫链的蒙特卡洛方法，并使计算时间大大缩短。本文用贝叶斯法对硬蜱属(Ixodes)19个种的线粒体16S rDNA片段进行了系统进化分析。从总体上看，分析结果与现有的基于形态学的分类体系基本吻合。但与现存的假说相反，莱姆病的主要宿主蓖籽硬蜱复合种组并非单系。通过比较贝叶斯法与其它三种方法的结果，我们认为贝叶斯法是一种系统进化分析的好方法，它既能根据分子进化的现有理论和各种模型用概率重建系统进化关系，又克服了最大似然法计算速度慢、不适用于大数据集样本的缺陷。贝叶斯法根据后验概率直观地表示系统进化关系的分析结果，不需要用自引导法进行检验。可以预料，贝叶斯法将会被广泛地应用到系统进化分析上[动物学报49(3)：380—388，2003]。

Application of Bayesian method for inferring phylogenetic relationships of ticks within the genus Ixodes

Three major methods, Neighbor-Joining (NJ), Maximum Parsimony (MP), and Maximum Likelihood (ML), are widely used for phylogenetic reconstruction. ML method has many advantages over NJ and MP but less practical. It is almost impossible to apply ML method to a sequence dataset that contains more than 25 operational taxonomic units (OUT) due to the enormous computational time required and the shortage of computer memories. The Bayesian method is a recently developed method that is similar to ML method in principle, but the computation time is reduced dramatically. Bayesian method inferences tree based on the notion of posterior probabilities: probabilities that are estimated, based on some model (prior expectations) using Markov chain Monte Carlo method. Instead of seeking the tree that maximizes the likelihood of observing the data, it seeks those trees with the greatest likelihood of given the data. Bayesian method produces a set of trees of roughly equal likelihood. To introduce and promote Bayesian method in Chin, we applied this method to the phylogenetic analysis of 19 tick species in the genus Ixodes. The inferred phylogeny is generally congruent with that based on morphology. However, in contrast to the existing hypothesis, 16S rDNA data indicated that I. ricinus species complex is not a monophyletic group. The results from Bayesian method were also similar to that of NJ and MP methods but with a few exceptions. There is no need to conduct bootstrap analysis to evaluate branch support because divergences on the tree are represented by the posterior probabilities.