The Performance of Iterative Search Strategies for Maximum-Likelihood Estimation of Phylogeny from DNA Sequences

从DNA序列数据中构建或估计系统发育分支结构（树），无论是生物体还是其组成基因，在广泛的生物学研究中越来越重要。最大似然（ML）估计方法也越来越多地用于构建系统发育树，因为这些方法具有一些理想的统计特性，包括修改被认为会影响DNA碱基变化速率和模式的核苷酸取代参数的能力。描述核苷酸取代过程的现实模型现在可用于计算似然值；例如，所有六种可逆核苷酸替代类型都可以允许以独特的相对速率进化，碱基频率可以偏离相等，不同的位点（和不同的密码子位置）可以允许以不同的速率进化。因此，目前使用的最复杂的时间可逆模型除了计算被检查树的分支长度外，还包括对十个参数的估计来描述核苷酸替代过程。这种复杂性的代价是计算最大似然得分所需的纯粹计算时间，这对于几乎所有在分析中处理20多个物种的研究人员来说都是令人望而却步的。然而，基因测序和基因组测序项目正在为更多的物种进行，对这些庞大数据集的分析需要最大似然搜索方法的改进。爱达荷大学的Jack Sullivan博士和他的同事David Swofford博士正在探索使用广泛可用的软件包PAUP*并行处理ML搜索的选择，以便对大型数据集中大量可能的树进行迭代搜索。一个多步骤的方法将从涉及少量分类群的已发表的系统发育数据集和树开始，然后转移到更多的分类群，然后添加针对50个左右分类群的目标数据集的迭代搜索策略的模拟研究。总体目标是改进当前的ML方法，以找到最优树结构和拓扑结构，分支长度和模型参数的最优组合，以在一个群体的进化变化中进行核苷酸取代。

9974124Sullivan and Swofford The construction or estimation of phylogenetic branching structure (trees) from DNA sequence data, whether for organisms or their constituent genes, is increasingly important in a wide array of biological studies. Maximum-likelihood (ML) estimation methods are also increasingly being used to construct phylogenetic trees because of several desirable statistical properties of the methods, including the capacity to modify parameters of nucleotide substitution that are thought to influence the rates and patterns of DNA base changes. Realistic models describing the process of nucleotide substitution are now available for use in calculating likelihood values; for example, all six reversible nucleotide substitution types can be allowed to evolve at a unique relative rate, base frequencies can deviate from equality, and different sites (and different codon positions) can be allowed to evolve at different rates. The most complex time-reversible models currently in use thus incorporate the estimation of ten parameters to describe the process of nucleotide substitution, in addition to the calculation of branch lengths for the tree under examaination. The penalty in this complexity is the sheer computational time required to calculate the maximum likelihood score, which becomes prohibitive for nearly all researchers working with more than 20 species in an analysis. Yet, gene-sequencing and genomic sequencing projects are underway for vastly more species, and the analysis of these huge datasets requires improvement in maximum-likelihood search methods. Dr. Jack Sullivan at the University of Idaho, working with his colleague Dr. David Swofford, is exploring the option of parallel processing of ML searches using the widely available software package PAUP*, in order to conduct iterative searches on the huge numbers of possible trees from large datasets. A multi-step approach will start with published datasets and trees for phylogenies involving small numbers of taxa, then move to larger numbers of taxa, and then add simulation studies on the iterative-search strategy focused on a target dataset of 50 or so taxa. The general goal is to improve upon current ML methods for finding both the optimum tree structure and the optimal combination of topology, branch lengths, and parameters of the model for nucleotide substitution during evolutionary change of a group.