Collaborative Research: Bayesian Model Checking for Phylogenetics in the Post-Genomic Era

物种和基因的进化关系图（系统发生树）广泛应用于生物研究，包括医学、流行病学、法医学、保护、进化生物学和农业等领域。该研究项目将探索新的思路并开发新的软件工具，以提高确定系统发育关系的准确性；通过这种方式，研究将有助于提高对广泛的科学学科和实际应用的理解和决策。这项研究的结果将广泛传播，包括面对面和在线培训机会，使有关学科的研究人员熟悉这些新开发的基于计算机的分析工具。此外，研究活动将涉及路易斯安那州立大学（LSU）和夏威夷大学马诺阿分校的一名博士后学者、一名研究生和几名本科生的参与和培训。该项目将被纳入路易斯安那州立大学的一系列研讨会，重点是提高本科生对计算生物学的认识。系统发育树现在通常是从庞大的基因组规模的数据集推断出来的，揭示了在不同位点上明显的系统发育信号的广泛变化。然而，目前还没有通用的工具来客观和定量地评估这种变化有多少是由于生物过程造成的，有多少是由方法错误造成的。区分真正的变异和错误是这个项目要研究的问题，因为解决这个问题对于有力地解决生命之树和理解基因组进化至关重要。这项工作的目标是为研究人员提供工具，以识别和避免系统发育推断不可靠的情况。这些工具将在开源软件（RevBayes和R）中实现，并且可以很容易地扩展到本项目之外的许多类型的系统发育推断。本研究将采用贝叶斯后验预测实现现有的统计方法套件，以严格评估系统发育模型与进化数据的绝对拟合，以及这种拟合如何影响推断的可靠性。模拟比较替代模型的性能将集中在三种类型的推断上：(i)单个基因树的估计，（ii）来自许多基因的物种树的估计，（iii）连续性状的比较分析。这些方法将应用于典型的经验问题，包括使用几个最近发表的基因组规模数据集在羊膜中放置海龟。这些数据包含了关于海龟安置的系统发育信号的惊人和巨大的异质性，因此形成了一个很好的案例研究。

Diagrams of evolutionary relationships (phylogenetic trees) for species and genes are widely employed in biological research, including the fields of medicine, epidemiology, forensics, conservation, evolutionary biology and agriculture. This research project will explore new ideas and develop new software tools to improve the accuracy by which phylogenetic relationships are determined; in this way the research will contribute to improved understanding and decision-making for a broad range of scientific disciplines and practical applications. Results from this research will be broadly disseminated, including in-person and online training opportunities to familiarize researchers in the relevant disciplines with these newly developed computer-based analytical tools. Further, the research activities will involve the participation and training of a postdoctoral scholar, a graduate student, and several undergraduates at Louisiana State University (LSU) and the University of Hawaii at Manoa. This project will be incorporated into a seminar series at LSU focused on increasing awareness of computational biology among undergraduate students.Phylogenetic trees are now routinely inferred from enormous genome-scale data sets, revealing extensive variation in apparent phylogenetic signal across loci. However, no general tools currently exist to objectively and quantitatively assess how much of this variation is due to biological processes and how much is caused by methodological error. Distinguishing between true variation and error is the problem to be studied in this project, as resolving this issue is essential for robustly resolving the Tree of Life and for understanding genomic evolution. The goal of this work is to give researchers the tools to identify and avoid situations where phylogenetic inferences are unreliable. These tools will be implemented in open-source software (RevBayes and R), and will be easily extensible to many types of phylogenetic inference beyond those in this project. This research will implement suites of existing, alternative statistical approaches employing Bayesian posterior prediction to rigorously assess absolute fit of phylogenetic models to evolutionary data, and how this fit impacts the reliability of inference. Simulations comparing performance of alternative models will focus on three types of inferences: (i) estimation of individual gene trees, (ii) estimation of species trees from many genes, and (iii) comparative analysis of continuous traits. These approaches will be applied to exemplar empirical questions, including the placement of turtles among amniotes using several recently published genome-scale data sets. These data contain surprising and massive heterogeneity in phylogenetic signal regarding the placement of turtles, and thus form an excellent case study.