Bayesian Dynamical Modeling of Microbial Communities

项目概要/摘要 微生物群落及其宿主在许多应用中发挥着关键作用，包括保护人类或植物 防治疾病或开发下一代生物燃料和生物修复系统， 可持续增长深入了解这些系统的基础生物学是解决问题的关键。 挖掘他们的潜力高通量多组学技术的进展，如宏基因组学，元转录， 组学，外代谢组学和蛋白质组学，使我们能够捕捉这些复杂的生物学的多个快照， 过程一次。这些快照创建组学特征的大规模高维数据集（例如，中- 微生物物种、微生物基因、蛋白质和小分子）。降低的成本也让研究人员 收集更多的多组学时间序列数据。这些时间分辨的多组学特征可以一起提供 生物学过程及其潜在活动的全面图景。 这些精心设计的多组学研究尚未被充分分析，主要是由于 缺乏进行这种分析所需的适当工具和注释数据库。例如，系统地 研究该纵向数据的时间分量以研究组学特征的时间动态 与疾病活动的关系是许多研究中未满足的需求。因此，迫切需要 统计工具，通过整合不同的数据类型和系统地 调查这个纵向数据的时间成分。 该项目的总体目标是开发有效的，可解释的，可扩展的工具，基于我们以前的 开发的信号模型，称为部分观测布尔动态系统（POBDS），以表征时间 通过多组学数据组成和捕获微生物群落的动态行为。原始 可按以下研究目标组织投稿： (i)在POBDS背景下开发新方法，能够对通过以下方法获得的多组学数据进行建模： 各种分子特征分析技术和各种疾病/领域。 (ii)开发贝叶斯优化框架，以实现网络的有效和可扩展重建 微生物群落的拓扑学（即，推断大量基因之间的相互作用类型， 细菌和微生物）。 (iii)开发贝叶斯强化学习扰动策略以减少所需的数据数量 用于建模/学习过程（克服不可识别性问题），并获得最丰富的信息 微生物群落的数据。 在这个项目中开发的所有工具将在一个用户友好的软件/工具免费访问其他 研究人员

Project Summary/Abstract Microbial communities and their hosts play a key role in many applications, including protecting humans or plants against diseases or developing the next generation of biofuels and biological remediation systems needed for sustainable growth. Gaining a deep understanding of the fundamental biology of these systems is the key to har- nessing their potential. Advances in high-throughput multi-omics techniques like metagenomics, metatranscrip- tomics, exometabolomics, and proteomics, allow us to capture multiple snapshots of these complex biological processes at once. These snapshots create large-scale high-dimensional datasets of omics features (e.g., mi- crobial species, microbial genes, proteins, and small molecules). The reduced cost has also allowed researchers to collect more multi-omics time-series data. These temporally resolved multi-omics features can together provide a comprehensive picture of biological processes and their underlying activities. These well-designed multi-omics studies have not been analyzed to their fullest potential yet, primarily due to the lack of appropriate tools and annotation databases required for such analyses. For example, systematically investigating the time component of this longitudinal data to investigate the temporal dynamics of omics features in relationship with disease activities is an unmet need in many studies. Therefore, there is a critical need for statistical tools to greatly improve research infrastructure by integrating different data types and systematically investigating the time component of this longitudinal data. This project's overarching goal is to develop efficient, interpretable, and scalable tools based on our previously developed signal model, called partially-observed Boolean dynamical systems (POBDS), to characterize the time component and capture the dynamical behavior of microbial communities through multi-omics data. The original contributions can be organized across the following research goals: (i) Developing novel methods in the POBDS context capable of modeling multi-omics data obtained through various molecular profiling technologies and various diseases/domains. (ii) Developing Bayesian optimization frameworks for the efficient and scalable reconstruction of the network topology of microbial communities (i.e., inferring the type of interactions between a large number of genes, bacteria, and microbes) through high dimensional multi-omics data. (iii) Developing Bayesian reinforcement learning perturbation policies to decrease the number of data required for the modeling/learning process (overcoming the non-identifiability issue) and acquire the most informative data in microbial communities. All the developed tools in this project will be presented in a user-friendly software/tool freely accessible to other researchers.