CRII: III: Using Genomic Context to Understand Evolutionary Histories of Individual Genes

进化树通过进化时间展示了生物体之间的关系，并且是许多生物医学应用的核心，例如理解蛋白质功能和分析人类微生物组。有趣的是，进化历史可以在整个基因组中发生变化，从而形成复杂的基因组景观。理解全基因组进化需要复杂的计算工具，这些工具可以分析跨越生命之树的物种快速增长的基因组序列。目前可用的推断这些复杂进化历史的方法存在严重的局限性，无论是在准确性方面还是在大型数据集的可扩展性方面。该项目将开发新的计算方法，用于重建基因组单个片段的进化树，同时考虑它们与基因组其他部分的联系。开发的计算方法将公开提供给进化生物学家社区，并将有助于改善各种下游生物医学分析。为了促进采用，PI将培训对新方法感兴趣的生物学家。除了影响研究界之外，PI还将为加州大学圣地亚哥分校新的有前途的多样性倡议做出贡献，以吸引URM/女本科生参与生物学和计算机之间令人兴奋的重叠研究项目。“基因树”显示了一小部分基因组的进化史，而准确重建每个基因树可能很困难，因为一小部分可能提供的数据很少，而且只有微弱的信号。改进基因树估计的一个原则性方法是考虑基因组不同部分之间的联系。然而，这种共同估计方法的计算局限性使研究人员只有一种可行的选择：独立地推断每个片段的基因树，而忽略了潜在的全基因组进化模式。这个项目将开发新的可扩展的统计方法来推断基因树，给出基因组序列数据和进化历史的全基因组描述，即所谓的物种树。重点将放在基因树的一个特定的概率模型上，即多物种聚合模型，它将等位基因分化的种群水平模式与物种水平的历史联系起来。该模型下的可扩展基因树推理仍然是一个研究较少的问题。为了开发可扩展的方法来计算和优化多物种聚结模型的似然分数，该项目将使用分而治之和动态规划等算法技术。其他将探索的方法包括设计简化的优化分数，比完全似然更容易计算，并开发新的似然函数近似值。如果成功，该项目将为多物种聚结模型下的物种树推断基因树提供开创性的可扩展解决方案，这种方法将对下游生物分析产生实质性的积极影响。此外，该项目可以为更复杂的推理问题的可扩展解决方案铺平道路，其目标是在一次分析中共同估计大量的基因树和物种树。欲了解更多信息，请参阅项目网页：http://eceweb.ucsd.edu/~smirarab/2016/01/27/genetrees.html

Evolutionary trees show relationships between organisms through evolutionary time, and are central to many biomedical applications such as understanding protein function and analyzing the human microbiome. Interestingly, evolutionary histories can change across the genome, creating a complex genomic landscape. Understanding genome-wide evolution requires sophisticated computational tools that can analyze the rapidly growing set of genomes sequenced from species spanning the tree of life. Currently available methods for inferring these complex evolutionary histories have serious limitations, either in terms of accuracy or scalability to large datasets. This project will develop new computational methods for reconstructing evolutionary trees for individual segments of the genome while considering their connections to other parts of the genome. The computational methods developed will be made publicly available to the community of evolutionary biologists, and will help improve various downstream biomedical analyses. To facilitate adoption, the PI will train biologists interested in the new methods. Beyond impacting the research community, the PI will contribute to new promising diversity initiatives at UC San Diego to engage URM/Women undergraduate students in research projects defined in the exciting overlap between biology and computing.A "gene tree" shows the evolutionary history for a small segment of the genome, and accurate reconstruction of each gene tree can be difficult because a small segment may provide little data and only a weak signal. A principled approach for improving gene tree estimation is to consider the connection between different parts of the genome. However, computational limitations of such co-estimation approaches have left researchers with only one feasible option: to infer gene trees independently for each segment, ignoring the underlying genome-wide patterns of evolution. This project will develop new scalable statistical methods for inferring gene trees given genomic sequence data and a genome-wide description of the evolutionary history, the so-called species tree. The focus will be on a specific probabilistic model of gene trees, the multi-species coalescent model, that links population level patterns of allele differentiation to species level histories. Scalable gene tree inference under this model remains a poorly studied problem. To develop scalable methods for calculating and optimizing likelihood scores with respect to the multi-species coalescent model, the project will use algorithmic techniques such as divide-and-conquer and dynamic programming. Other approaches that will be explored include designing simplified optimization scores that are easier to compute than the full likelihood and developing new approximations to the likelihood function. If successful, the project will produce pioneering scalable solutions for inferring gene trees given the species tree under the multi-species coalescent model, and such a method will have substantial positive impact on downstream biological analyses. Moreover, this project can pave the way for scalable solutions to an even more complex inference problem, where the goal is to co-estimate a large number of gene trees and the species tree in one analysis. For further information see the project web page: http://eceweb.ucsd.edu/~smirarab/2016/01/27/genetrees.html