Novel statistical methods for controlled variable selection of microbiome data

项目概要/摘要 科学界越来越重视微生物群落的重要作用， 在许多疾病和健康状况中发挥作用。微生物群落的结构（例如，相对 不同分类群和微生物网络/相互作用的丰度）会随着许多 环境和宿主因素。科学研究微生物如何相互作用， 环境及其宿主之间的相互作用可以揭示我们对人类免疫缺陷的潜在生物学机制的理解。 与微生物组相关的疾病和健康状况。尽管有着令人难以置信的研究兴趣和 通过尖端技术（16 S rRNA基因）的创新使用， 测序、鸟枪宏基因组学测序和代谢组学），统计工具仍然不足 可以完全处理微生物组数据的复杂性，包括高维，系统发育， 相关性、相对较小的样本量、组成限制等。该提案的主要目标是 开发统计功能强大和计算效率高的方法来解决这些挑战， 微生物组数据。特别是，这项研究将应用于高通量微生物组数据，并导致 新的统计控制变量选择方法，1）选择真正 在预先规定的错误发现率（FDR）下与疾病相关结局相关，其中 结果可以是感兴趣的单一疾病结果，也可以是多变量的，例如多个继发性 与疾病相关的表型;和B）鉴定与疾病相关的分类群和分类群-代谢物相互作用 在一定的FDR阈值下的疾病结果。我们提出的方法是创新的，因为它可以 两者都选择重要的分类群特征或分类群-代谢物相互作用，并具有受控的FDR， 大大提高了微生物组关联研究中发现结果的可重复性和可靠性。 增强的分类群选择将进一步促进下游基于实验室的功能研究， 最终可能改善许多健康问题的预防、检测、治疗和监测， 和疾病状况之间的关系。完成这项建议也将有助于弥合 在微生物组研究的研究兴趣和缺乏分析工具之间的差距。在 除了在同行评审的期刊上发表，我们还将通过会议传播我们的成果 和开放源码软件，可以免费提供给更广泛的科学界。所提出的方法是 对于更好地理解微生物组机制沿着与宿主基因组的相互作用至关重要， 代谢组学在某些疾病的病理学中起重要作用，这些疾病对人类健康至关重要。

Project Summary/Abstract The scientific community is increasingly appreciative of the important role that the microbiome community plays in many diseases and health conditions. The structure of the microbiome community (e.g., relative abundances of different taxa and microbial network/interactions) is subject to change in response to many environment and host factors. Scientific investigation of how microbiome interact with each other, with their environment and with their host can shed light on our understanding of the underlying biological mechanism of microbiome-related disease and health conditions. Despite the incredible amount of research interest and availability of massive data through the innovative use of cutting-edge techniques (16S rRNA gene sequencing, shotgun metagenomics sequencing and metabolomics), there are still insufficient statistical tools that can fully handle the complexity of microbiome data, including the high-dimensionality, phylogenetic relatedness, relatively small sample size, compositional constraint and others. The main goal of this proposal is to develop statistically powerful and computationally efficient methods to address these challenges in analyzing microbiome data. In particular, this research will be applied to high-throughput microbiome data and lead to new statistical controlled variable selection methods that 1) select a subgroup of taxa that are genuinely associated with disease-related outcomes under a pre-specified false discovery rate (FDR), where the outcomes can be either a single disease outcome of interest or multivariate such as multiple secondary phenotypes related to the disease; and b) identify taxa and taxa-metabolite interactions that are associated with a disease outcome under a certain FDR threshold. Our proposed methods are innovative in that it can both select important taxa features or taxa-metabolites interactions and have the FDR being controlled, which largely enhances the reproducibility and reliability of the discovery results in microbiome association studies. The enhanced taxa selection would further facilitate downstream laboratory-based functional studies, eventually leading to potential improvements in prevention, detection, treatment and monitoring of many health and disease conditions from a microbiome's perspective. Completion of this proposal will also help bridging the gap between the burgeoning research interest in microbiome studies and the lack of analytical tools. In addition to publication in peer-reviewed journals, we will make our results disseminated through conferences and open-source software that is freely available to the wider scientific community. The proposed methods are essential for improved understanding of microbiome mechanism along with its interaction with host genome or metabolome in the pathology of certain diseases, which are of central importance to human health.