Sequence data and the ecology of pathogens: phylogeny and beyond

该建议旨在通过以一种新的方式理解病原体的进化来提高我们推断形成病原体进化的生态过程的能力。了解病原体传播的生态学是很重要的。事实上，生态学思想在理解病原体方面有很大的帮助：生态学家习惯于复杂的数据集，没有机会进行真正的控制实验，生态学概念，如竞争和竞争排斥，生态位适应和栖息地过滤越来越多地成为理解病原体进化的范式。一个与生态学思想特别相关的问题的例子是，一些病原体如何以及为什么迅速发展出广泛的耐药性，而另一些病原体则保持耐药菌株和敏感菌株的长期共存。原则上，病原体传播包含大量信息，既有关于某些菌株或序列起源的细节，也有关于形成何时、何地以及哪些病原体菌株能够传播的一般潜在过程的信息。数学家们已经开发出工具来创建最大似然系统发育树，代表数据集的估计祖先模式，以及同时推断遗传和人口统计历史的工具。然而，现有的工具没有提供系统的方法来推断形成病原体的传播或进化的生态环境。此外，目前的方法使用非常有限的指标来评估遗传学，因此，在某种程度上，确实存在使用遗传数据来了解流行病学和生态过程的方法，这些方法基于遗传数据中非常丰富的信息。这里提出的工作旨在填补正在收集的病原体基因组的丰富数据集与我们分析它们的能力之间的差距。我将首先开发一套方法来总结系统发育树，考虑到树的拓扑结构。现有的方法识别树的概率与它的分支时间的概率，忽略了树的拓扑结构。由于影响树木的其他因素（随机传播和突变，选择等），开发信息性措施可能具有挑战性。对于每一个新的摘要，我的目标是找到它在从树空间均匀绘制的随机树上的分布，以确定给定树的稀有程度。在第二阶段的工作，我的目标是改善推理的基础生态过程塑造病原体的进化，通过更好地了解什么功能的系统发育树（包括新的摘要措施在第一阶段开发），使他们能够解释观察到的数据。从遗传学中推断生态过程将带来一些与推断人口统计学中出现的相同的挑战，其中之一是n叶上可能的树的数量太高，无法对所有这些树进行求和。然而，这样的总和是基于可能性的推理的核心。我建议使用在建议的第一阶段中确定的特征，以及对给定树G，L（D）的数据D的可能性如何的更好理解|G）分布在树空间上，以简化和并改善推理。

This proposal aims to improve our ability to infer the ecological processes shaping a pathogen's evolution by understanding pathogen phylogenies in a novel way. It is important to understand the ecology of pathogen spread. Indeed, ecological ideas have much to offer in understanding pathogens in particular: ecologists are accustomed to complex datasets without the opportunity for truly controlled experiments, and ecological concepts such as competition and competitive exclusion, niche adaptation, and habitat filtering are increasingly the paradigm of choice for understanding pathogen evolution. An example of a question for which ecological ideas are particularly relevant is that of how and why some pathogens evolve widespread drug resistance rapidly while others maintain long-term coexistence of resistant and sensitive strains. Pathogen phylogenies contain a lot of information, in principle, both about the specifics of where certain strains or sequences originate and about the general underlying processes shaping when, where, and which pathogen strains are able to spread. Mathematicians have developed tools to create maximum-likelihood phylogenetic trees representing the estimated ancestral patterns of a dataset, as well as tools to simultaneously infer a phylogeny and the population's demographic history. However, existing tools offer no systematic approaches to infer the ecological context shaping a pathogen's spread or evolution. In addition, current methodologies use very limited metrics to assess phylogenies, so to the extent that approaches do exist to use genetic data to understand epidemiological and ecological processes, these are based on very little of the rich information in genetic data. The work proposed here aims to fill the gap between the rich datasets of pathogen genomes being gathered and our ability to analyse them. I will first develop a suite of ways to summarise phylogenetic trees, taking the topology of the tree into account. Existing methods identify the probability of a tree with the probabilities of its branching times, neglecting the tree's topology. Developing informative measures is likely to be challenging because of the other factors that affect trees (stochastic transmission and mutation, selection, and others). For each new summary I aim to find its distribution on random trees drawn uniformly from tree space, to determine how rare a given tree is. In the second stage of the work I aim to improve inference of the underlying ecological processes shaping pathogen evolution, by better understanding what features of phylogenetic trees (including the novel summary measures developed in the first stage) make them able to account for observed data. Inference of ecological processes from phylogenies will carry some of the same challenges that occur in the inference of population demographics, one of these being that the number of possible trees on n leaves is too high for summing over all such trees to be feasible. Yet such sums are at the heart of likelihood-based inference. I propose to use the features identified in the first stage of the proposal, together with an improved understanding of how the likelihood of the data D given a tree G, L(D|G), is distributed over tree space, to simplify the sum and improve inference.