Estimating the Bayesian Phylogenetic Information Content of Systematic Data

这项研究将开发和测试新的分析方法，以估计用于确定物种之间的系谱关系(系谱)的生物数据集的质量和信息含量。在生物学的许多领域，从识别新出现的病原体到研究进化变化的机制和自然界物种群落的功能，系统发育都是至关重要的。这项研究项目中将要开发的方法将在科学期刊上公布，并将以免费、开放源码的软件提供，使研究人员能够将新方法应用于他们自己的数据。该项目还将促进生物学和统计学方面的博士后助理、研究生和教职员工的跨学科培训。高中生将参与开发移动应用程序，这些应用程序将为科学界提供有用的免费工具。贝叶斯统计框架广泛应用于系统发育和分子进化；然而，估计数据信息量的方法仍然发展不足。其中一个原因可能是这样一个事实，即估计Kullback-Leibler(KL)发散需要边缘似然和拓扑后验概率，并且直到最近才实现了估计这些量的准确方法。相应地，这项研究的主要目标是：(1)评估基于KL的信息量测量的效用，以回答对系统学家来说很重要的各种问题，包括：(A)数据集中存在多少关于树拓扑的信息？(B)能否将拓扑信息与有关替代模型参数的信息分开？(C)一个数据子集与不同的数据子集相比有多少信息？(D)两个数据子集是否包含相互冲突的系统发育信息？(E)有多少关于特定模型参数的信息(例如，感兴趣的发散时间)？(F)有多少信息可用于解析特定的分支？(2)探索与信息内容有关的问题，例如：(A)饱和的拓扑信息内容定义；(B)为模型选择的目的估计变量树边缘似然的基于KL的方法；以及(C)多元分析。(3)在现有的贝叶斯系统发育软件中实现对信息量的KL估计，使其免费提供给系统学界。这些目标将使用变量树IDR方法来实现，以提供KL估计所需的所有边际似然的准确估计。最近发布的条件分支方法允许从样本条件分支频率估计树拓扑的准确后验概率。将利用计算机模拟实验和对有关真实世界数据集的分析来评估知识学习在衡量信息内容和为上述问题提供答案方面的有效性。

This research will develop and test new analytical methods for estimating the quality and information content of biological data sets used to determine genealogical relationships (phylogenies) among species. Phylogenies are crucial in many areas of biology, from identifying emerging pathogens to studying the mechanisms of evolutionary change and the functioning of communities of species in nature. The methods to be developed in this research project will be announced in scientific journals, and will be provided in free, open-source software enabling researchers to apply the new methods to their own data. This project will also facilitate interdisciplinary training of postdoctoral associates, graduate students and faculty in biology and statistics. High school students will be involved in developing mobile apps that will provide useful, free tools to the scientific community.The Bayesian statistical framework is widely used in phylogenetics and molecular evolution; however, the means for estimating the information content of data remains poorly developed. One reason for this may be the fact that marginal likelihoods and topology posterior probabilities are required for estimating Kullback-Leibler (KL) divergences, and accurate means of estimating these quantities have only recently been achieved. Correspondingly, primary objectives for this research are: (1) Evaluate the utility of KL-based information content measurement to answer a diversity of questions important to systematists, including: (a) How much information about tree topology is present in a data set? (b) Can topological information be separated from information about substitution model parameters? (c) How much information does one data subset have compared to a different data subset? (d) Do two data subsets contain conflicting phylogenetic information? (e) How much information is there about particular model parameters (e.g. a divergence time of interest)? (f) How much information is there for resolving particular clades? (2) Explore issues related to information content, such as: (a) a topological information content definition of saturation; (b) a KL-based method for estimating variable-tree marginal likelihoods for purposes of model selection; and (c) polytomy analyses. (3) Implement KL estimation of information content in existing Bayesian phylogenetics software to make it freely available to the systematics community. These objectives will be pursued using the variable tree IDR method to provide accurate estimates of all marginal likelihoods needed for KL estimation. The recently published conditional clade method allows accurate posterior probabilities of tree topologies to be estimated from sample conditional clade frequencies. Computer simulation experiments and analyses of relevant real world data sets will be used to evaluate the effectiveness of KL in measuring information content and in providing answers to the questions posed above.