Novel Statistical Methods in Analyzing Microbiome Data for Longitudinal Study

标题：分析纵向研究微生物组数据的新统计方法 摘要： 最近的研究表明，微生物组的变化可能对健康产生相当大的影响， 如营养不良、哮喘、肥胖、糖尿病和其他疾病。这进一步促进了实质性的兴趣 从基础和临床角度来看， 旨在探索微生物组如何影响健康和疾病的机制。但特别 高维组成微生物组数据结构和特征使有效分析复杂化 微生物数据。具体而言：1）微生物组数据是组成性的; 2）微生物组数据是高维的; 3)细菌分类群通过系统发育树在进化上相关;以及4）微生物组组成通常 量化为稀疏组成数据向量（具有单位和的比例的稀疏向量）。局限性 适当的统计方法来分析这种独特的高维数据阻碍了我们做出推断的能力， 得出关于微生物组在人类健康和疾病中的作用的结论。受到挑战的激励 我们在对微生物组改变的影响进行纵向微生物组合作研究时， 就第一型糖尿病的发展，我们建议达到以下的具体目标：（1）设计一个 比较各组之间微生物组的时间变化的框架;（2）制定纵向 在微生物诱导的复杂性状/疾病（糖尿病）研究中推断因果关系的中介模型;（3）开发 一个统一的，强大的，强大的统计框架，以测试细菌类群之间的关联， 微生物组研究中的疾病/特征;以及4）开发，分发和支持免费提供的软件 我们开发的方法的包。将通过分析方法对这些方法进行评价， 计算机模拟和应用于多个真实的数据集。该应用程序的长期目标是 开发和实施新的统计方法来研究微生物组组成的动态， 确定影响对复杂性状敏感性的关键细菌物种，应用这些方法促进 研究，并将这些工具传播给一般研究界。

Title: Novel Statistical Methods in Analyzing Microbiome Data for Longitudinal Study Abstract: Recent research demonstrates that changes in the microbiome can have considerable health implications such as malnutrition, asthma, obesity, diabetes, and other conditions. This has further promoted substantial interest in the microbiome from both basic and clinical perspectives, and prospective longitudinal studies have been conducted to probe the mechanisms on how the microbiome affects health and disease. However, the special structure and characteristics of high-dimensional compositional microbiome data complicate effective analysis of microbiome data. In particular: 1) microbiome data is compositional; 2) microbiome data is high dimensional; 3) bacterial taxa are related evolutionarily by a phylogenetic tree; and 4) microbiome compositions are often quantified as sparse compositional data vectors (sparse vectors of proportions with unit sum). Limitations of proper statistical methods to analyze this unique high dimensional data hinder our ability to make inferences or draw conclusions about the role of the microbiome in human health and disease. Motivated by the challenges we have encountered during collaborative longitudinal microbiome studies on the effect of altered microbiome on the development of Type I Diabetes, we propose to accomplish the following specific aims: (1) to design a framework to compare the temporal changes of microbiome between groups; (2) to develop longitudinal mediation models to infer causality in microbe-induced complex trait/disease (diabetes) studies; (3) to develop a unified, powerful, and robust statistical framework to test the association between bacteria taxa and diseases/traits in microbiome studies; and 4) to develop, distribute and support the freely available software packages for the methods we develop. The methods will be evaluated through analytical approaches, computer simulations, and applications to multiple real datasets. The long-term goals of this application are to develop and implement novel statistical methods to study the dynamics of the microbiome composition, to identify key bacterial species that affect susceptibility to complex traits, to apply these methods to facilitate ongoing studies, and to disseminate these tools to the general research community.