Integrative mutation-selection models of protein-coding sequence evolution

目的设计简单易用的计算方法，实现丰富的分子进化概率系统发育模型。我们的建模目标沿着两个主要轴线运行：1)在基于统计健全的人口遗传学的公式中纳入更大的现实主义，以定义相关的零模型；2)开发替代模型，测试偏离我们日益现实的零模型的情况。沿着第一个轴，我们以前已经建立了蛋白质编码DNA序列的近中性进化的基本零模型。我们将通过努力捕捉上下文相关的突变特征、由于蛋白质结构限制而引起的跨位点相关性以及整个系统发育过程中不同的种群大小，来寻求改善我们空模型的结构。沿着我们建模目标的第二个轴，我们将专注于测量与近中性零模型的背离，这些背离导致自适应进化的特定地点签名。科学方法尽管上下文相关的突变模型或结构约束模型用传统方法在计算上很难处理，但我们将寻求两种不同的系统来实现这些模型：近似贝叶斯计算(ABC)和基于数据增强的马尔可夫链蒙特卡罗方法，我们已经使用过这两种方法。我们将使用模拟来研究模型的性质和计算技术。我们还将根据实际数据，使用后验预测方法来评估替代模型的性能。最后，我们将在大量数据上应用模型，以获得对它们提供的衡量标准的经验性见解。新颖性和预期意义我们过去在丰富的进化模型及其实现方面的工作对我们理解生物之间的进化关系产生了重大影响，例如为建立现代动物树做出了贡献。还有更多的工作要做，才能设计出承认我们对分子进化生物学的基本理解的模型。尽管潜在的理论发展可以追溯到20世纪初，但整合了这一理论的分子进化机制模型直到最近才获得吸引力。我们在这一领域的研究工作最有特色的是，他们专注于改进的零模型。我们的模型的一些应用表明，它们可以进一步改进系统发育推理本身，但更根本的是，它们可以帮助揭示进化过程的潜在特征。我们的工作有助于一个广泛的建模项目，旨在将序列数据、蛋白质结构数据和其他形式的数据集成到一个基础良好的概率框架中，作为生物学推理的中央引擎。

Objectives We aim to devise tractable computational methods for implementing rich probabilistic phylogenetic models of molecular evolution. Our modeling objectives operate along two main axes: 1) the incorporation of greater realism within a statistically sound population-genetics-based formulation, so as to define a relevant null model; 2) the development of alternative models, that test for departures from our progressively more realistic null. Along the first axis, we have previously established a basic null model of nearly-neutral evolution for protein-coding DNA sequences. We will seek to improve the structure of our null model, by pursuing our efforts at capturing context-dependent mutational features, across-site dependence due to protein structure constraints, and variable population sizes across the branches of the phylogeny. Along the second axis of our modeling objectives, we will focus on measuring departures from the nearly-neutral null model that result in site-specific signatures of adaptive evolution. Scientific approach Although context-dependent mutation models, or structurally constrained models, are computationally intractable with conventional methods, we will pursue two different systems for implementing the models: Approximate Bayesian Computation (ABC) and data-augmentation-based Markov chain Monte Carlo methods, with which we have had previous success. We will use simulations to study the properties of the models and computational techniques. We will also use posterior predictive methods to assess the performance of alternative models in light of real data. Finally, we will apply models on large sets of data, to gain empirical insights on the measurements they provide. Novelty and expected significance Our past work on rich evolutionary models and their implementations has had a major impact on our understanding of the evolutionary relationships between organisms, such as contributing to establishing the modern tree of animals. Much more work remains to devise models that recognize our basic understanding of molecular evolutionary biology. In spite of the underlying theoretical developments having a history dating back to the early 20th century, mechanistic models of molecular evolution that integrate this theory have only recently gained traction. Our research efforts in this area are most distinctive in that they are focused on an improved null model. A few applications of our models suggest that they can further improve phylogenetic inference per se, but, more fundamentally, that they can help reveal underlying features of the evolutionary process. Our work contributes to a broad modeling project, aimed at integrating sequence data, protein structure data, and other forms of data, into a well-grounded probability framework, as a central engine to inference in biology.