Methodology Development and Implementation for Microbiome Sequencing Data: Hierarchical Modeling on Clustered Taxa Counts with Repeated Measures

下一代测序技术的进步使研究人员能够揭示微生物群落的广泛变异性及其与不同疾病的关系。因此，了解影响微生物组组成的环境和宿主遗传因素变得至关重要。然而，由于微生物组测序数据的复杂性，这一领域的可靠和强大的方法还不够发达，其中包括：a)微生物分类数据通常被分成可操作的分类单元(OTU)，这些计数往往高度倾斜、过度分散和零膨胀；b)分类等级聚类内的OTU计数通常高度相关，但这种多变量性质通常被忽视；c)研究设计往往涉及从相关家庭成员那里重复采取的措施，从而导致时间和家族相关性。在这项提案中，我将开发强大的生物信息学、统计学和计算方法来克服这些挑战。具体地说，我建议使用潜在变量(LV)方法在纵向家族研究框架内联合建模来自层级分类群的多个分类群。LV框架代表了该簇的基本概念特征，并解释了不同分类群之间的相互关系。为了解决分类群计数的过度分散和零膨胀的特征，我将对多变量OTU结果应用零膨胀模型和障碍模型。 LV推断将基于贝叶斯框架，使用马尔可夫链蒙特卡罗(MCMC)算法从后验分布中获得样本。将开发一种贝叶斯模型选择算法，以选择特定数据集的最佳模型。我将在遗传因素上加入降维方法，这样就可以从全基因组数据中识别出遗传关联信号。我还将在微生物组数据上探索基因-基因(GxG)和基因-环境(GxE)的相互作用。高效的计算算法将使用C++开发，计算软件将在一个用户友好的界面内实现，该界面将分发给微生物组研究社区。此外，将建立一个标准化的分析管道，用于微生物组数据的建模和分析，并通过模拟进行测试。还将提供基于理论推导和经验结果的样本量估计和功率分析，以便设计未来的研究。这项建议将有助于标准化和优化未来对微生物组测序研究中可改变的环境风险因素以及遗传因素的研究。这项研究计划将为加拿大和国际遗传学和计算生物学研究人员社区推进大规模微生物组测序分析技术。

The technological advances in next generation sequencing have enabled researchers to unveil the wide variability in microbial communities and their relationships with different diseases. Therefore, it is becoming critical to understand both environmental and host genetic factors that impact the composition of the microbiome. However, robust and powerful methods in this area are underdeveloped due to the complexity of microbiome sequencing data, which includes: a) microbial taxa data are usually grouped into operational taxonomic units (OTUs) and these counts are often highly skewed, over-dispersed, and zero inflated, b) OTU counts within a taxonomic hierarchical cluster are often highly correlated, but this multivariate nature is usually ignored, c) the study designs often involve repeated measures taken from related family members, thus inducing temporal and familial correlations. In this proposal, I will develop powerful bioinformatics, statistical, and computational methods to overcome these challenges. Specifically, I propose to use the latent variable (LV) methodology to jointly model multiple taxa from hierarchical taxonomic clusters within a longitudinal family study framework. The LV framework represents the underlying conceptual traits of the cluster and explains the correlations among different taxa. To address the over-dispersed and zero inflated features of the taxa counts, I will apply both zero-inflated and hurdle models on the multivariate OTU outcomes. The LV inference will be constructed based on a Bayesian framework with samplings from the posterior distribution obtained using Markov Chain Monte Carlo (MCMC) algorithms. A Bayesian model selection algorithm will be developed to choose the optimal models for a particular dataset. I will incorporate dimensionality reduction methodologies on the genetic factors so that the genetic association signals can be identified from genome-wide data. I will also explore gene-gene (GxG), and gene-environment (GxE) interactions on the microbiome data. High-efficiency computational algorithms will be developed using C++, and computational software will be implemented within a user-friendly interface which will be distributed to the microbiome research community. In addition, a standardized analytic pipeline for modeling and analysis of microbiome data will be constructed and tested by simulations. Sample size estimation and power analysis based on both theoretical deduction and empirical results will also be provided to allow design of future studies. This proposal will help standardize and optimize future research on modifiable environmental risk factors, as well as genetic factors, for microbiome sequencing studies. This research program will advance large-scale microbiome sequencing analytic technologies for Canadian and international genetics and computational biology researcher community.