MIM: Machine Learning, Systems Modeling, and Experimental Approaches to Understand the Universal Rules of Life of Microbiota Using Marine Time Series Data

了解微生物之间的关系，它们的基因和群落的功能，环境，以及这些关系如何反映生命的普遍规则仍然是微生物学的基本问题。研究小组正在利用宏基因组学数据，即整个群落的DNA含量，从海洋时间序列中解决这些问题。尽管宏基因组学已有十多年的历史，但宏基因组数据集的许多方面仍需要新的方法来提取有价值的信息。研究团队将应用最先进的综合机器学习、系统建模和实验方法来处理现有和新生成的时间序列宏基因组数据，以更好地了解微生物群落中的相互作用网络及其对微生物群落功能的影响，这对理解生态系统中元素的全球循环和能量处理具有重要意义。本提案中开发的机器学习和数学建模工具应该为宏基因组的基础分析提供新的途径。理论和计算工具也将直接有利于统计和机器学习社区的因果推理以及生态建模。最终，这些工具将使研究人员能够帮助揭示来自许多不同环境的微生物组内的普遍生命规则，包括存在于动物和植物中的微生物组。该项目将为博士后、研究生、本科生和高中生提供跨学科培训，重点关注数据科学、计算机科学、统计学、计算生物学、环境生物学和生态学等领域代表性不足的群体。项目期间开发的软件工具将分发给社区。在过去的二十年中，与南加州大学微生物观测站相关的圣佩德罗海洋时间（SPOT）系列在不同的时间尺度（每天、每周、每月和季节性）收集了时间序列标记基因、宏基因组和亚转录组数据，这些数据跨越了海洋的不同深度、位置和扰动（原始和污染）。利用丰富的可用时间序列数据，研究团队将开发机器学习，系统建模和实验方法，以了解微生物群落的普遍生命规则。该项目的具体目标是：(1)开发机器学习方法，通过宏基因组读取组装和分类，识别微生物群落内的所有微生物，无论是已知的还是新的微生物，以及病毒和质粒等可移动遗传元件的宿主；(2)进一步研究具有仿制品错误发现控制的格兰杰图形模型，将所得计算工具应用于SPOT数据，以确定已知微生物基因组之间的因果关系；宏基因组、组装基因组与环境因素。(3)基于前两个目标构建的因果网络，建立驱动生物丰度和群落结构的机制模型，如竞争、交叉摄食、病毒-宿主相互作用、放牧和物理运输，并建立应用于多样化和未来生态系统的预测框架。(4)利用邻近结扎实验以及微生物群落的动态和新特性，对预测的病毒-宿主相互作用进行实验验证。用户友好的软件包自动化分析宏基因组数据的程序将被开发。本研究的共同资助由生物海洋学和数学生物学项目提供。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Understanding the relationships among microbial organisms, the functioning of their genes and communities, the environment, and how these relationships reflect universal rules of life remain essential problems in microbiology. The team of investigators are leveraging metagenomics data, the collective DNA content of the entire community, from a marine time series to address these questions. Although the science of metagenomics is over a decade old, there are still many aspects of metagenomic datasets that require new approaches to extract valuable information. The research team will apply state-of-the-art integrative machine learning, systems modeling and experimental approaches to existing and newly generated time series metagenomic data to better understand the interaction networks in microbial communities and their impacts on microbial community function, which have major implications for understanding the global cycling of elements and processing of energy in ecosystems. The machine learning and mathematical modeling tools developed in this proposal should provide new avenues for fundamental analysis of metagenomes. The theory and computational tools will also directly benefit both the statistical and machine learning community on causal inference as well as ecological modeling. Ultimately, these tools will enable investigators to help uncover the universal rules of life within microbiomes from many different environments, including those present in animals and plants. The project will provide interdisciplinary training for postdoctoral fellows, graduate, undergraduate and high school students with emphasis on underrepresented groups in data science, computer science, statistics, computational biology, environmental biology and ecology. Software tools developed during the project will be disseminated to the community.Over the past two decades, the San Pedro Ocean Time (SPOT) Series associated with University of Southern California Microbial Observatory has collected time series marker gene, metagenomic, and metatranscriptomic data at different time scales (daily, weekly, monthly, and seasonally) across various depths, locations and perturbations (pristine and polluted) in the ocean. With the rich available time series data, the research team will develop machine learning, systems modeling, and experimental approaches to understand the universal rules of life of microbial communities. The specific aims of this project are to (1) develop machine learning approaches to identify all microbes, known or novel, within the microbial communities and also host of mobile genetic elements, such as viruses and plasmids, through metagenomic read assembly and binning, (2) further investigate the Granger graphical models with knockoff false discovery control, apply the resulting computational tools to the SPOT data to identify causal relationships among the known microbial genomes, metagenome assembled genomes, and environmental factors. (3) based on the causal networks constructed from the first two aims, develop mechanistic models driving organism abundances and community structure, such as competition, cross-feeding, virus-host interactions, grazing and physical transport, and develop a predictive framework for application to diverse and future ecosystems. (4) experimentally validate the predicted virus-host interactions using proximity-ligation experiments and the dynamics and emerging properties of the microbial communities. User-friendly software packages to automate the procedures for analyzing metagenomic data will be developed. Co-funding for this research was provided by the Biological Oceanography and Mathematical Biology programs.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.