Novel Bayesian Inference for Microbial Community Studies Using High-Throughput Sequencing Data

该项目的主要目标是开发计算高效的统计方法来分析大型复杂的宏基因组学数据，并对未来的微生物群落研究产生广泛的影响。特别是，通过精心构建的贝叶斯层次模型，将样品和物种之间的信息彻底结合起来，可以更好地理解环境和生物因素对微生物群落的影响。通过与加州大学圣克鲁斯分校海洋科学研究人员的密切合作，PI的研究结果将被整合到生物学中。PI将通过用户友好的软件在网上传播计算工具。虽然开发的模型是为了解决微生物群落研究中的问题而构建的，但它们一般适用于宏基因组学领域以外的应用。该项目将为研究生提供良好的培训机会，在贝叶斯建模的前沿方法方面开展方法研究，从实际重要和科学相关的应用领域出发。微生物群落与环境和宿主生物相互作用。科学研究微生物组如何与宿主、与环境以及彼此之间相互作用，对于更好地了解它们的功能作用（如对宿主疾病风险易感性的影响）具有重要意义。在微生物群落研究中，高通量测序技术的最新突破允许对环境样本中的微生物组进行深度分析，并为复杂测序数据的分析带来统计学上的新挑战。三个主要目标是：(1)发展贝叶斯非参数（NP）方法来概率表征微生物物种丰度水平的分布；(2)建立贝叶斯回归模型，研究物种丰度与协变量的关系；(3)建立贝叶斯图解方法来模拟物种间的相互作用。不同于将单个物种视为单独的结果，图将被用于表示概率模型中物种之间的依赖关系，并识别物种之间的重要关系。这将通过诚实的不确定度量化，为物种的共发生和共排除提供更完善的推断，并有助于更好地理解微生物的相互作用。此外，PI将开发可扩展和高效的计算算法，对大型测序数据进行后验推理。

The main goal of this project is to develop computationally efficient statistical methods for analyzing large complex metagenomics data and generate a broad impact on future microbial community studies. In particular, by thoroughly combining information across samples and species through a carefully constructed Bayesian hierarchical model, better understanding of influences of environmental and biological factors in microbial communities will be made possible. The PI's findings will be integrated into biology through close collaboration with researchers in Ocean Sciences at University of California Santa Cruz. The PI will disseminate computational tools online via user-friendly software. The developed models, although motivated and constructed to address the problems in microbial community studies, are applicable in general to applications outside the metagenomic field. The project will provide graduate students excellent training opportunities in cutting-edge methods on Bayesian modeling to develop methodological research motivated from practically important and scientifically relevant application areas. Microbial communities interact with the environments and host organisms. Scientific investigations of how microbiome interacts with their host, with their environment, and with each other are of great importance to better understanding of their functional roles such as influences on disease risk susceptibility of the host. In microbial community studies, recent breakthroughs in high-throughput sequencing technology allow deep interrogation of microbiome taken from environmental samples and generate statistically new and interesting challenges for analysis of complex sequencing data. The three main objectives are: (1) to develop Bayesian nonparametric (NP) approaches to probabilistically characterize distributions of abundance levels of microbial species; (2) to develop Bayesian regression models to investigate association between abundance of species and covariates; and (3) to develop a Bayesian graphical method to model interaction of species. Instead of treating individual species as separate outcomes, graphs will be used to represent dependencies between species in a probabilistic model and to identify important relationships between species. This will provide more improved inference on the co-occurrence and co-exclusion of the species with honest uncertainty quantification and lead to greater understanding of the microbial interactions. In addition, the PI will develop scalable and efficient computational algorithms to carry out posterior inference for large sequencing data.