DMS/NIGMS 2: Bayesian Statistical Methods for Comprehensive Inferences on Microbial Community Dynamics Using High-Throughput Sequencing Data

We propose to develop flexible Bayesian statistical methods to gain a comprehensive understanding of microbial community dynamics using high-throughput sequencing data. The emergence of large-scale microbiome studies provides new opportunities for understanding how various microbial communities function and relate to their environment. However, the analytical methodology required to model complex microbiome data is still lacking. One of the key objectives is to develop a general method for inferring microbial community dynamics that vary with host and environmental factors. We also aim to extend this method to complex scenarios, such as longitudinal microbiome studies, which investigate the evolution of microbial communities, and multi-omics microbiome studies that integrate various types of omics data. Our proposed methods rigorously address the unique challenges of microbiome data analysis and achieve more accurate inferences about the underlying biological processes with honest uncertainty quantification. The proposed methods will provide an opportunity to attain a deeper understanding of the microbiome’s role, potentially paving the way for intervention strategies that enhance health and disease management. The proposed research involves synthesizing innovative concepts to tackle statistical challenges in microbiome data analysis within complex study settings, with a particular focus on multivariate count data presenting unique statistical complexities. The research agenda is broad and widely applicable, consisting of methodological development and theoretical examination of model properties, along with a challenging computational component aimed at achieving computational feasibility for big data. Our semiparametric methods offer significantly improved accuracy compared to existing methods. Our innovative approach to imposing a joint sparsity structure on the covariance matrix enhances the ability to infer microbial interactions. This approach improves robustness against large signals and reduces noise in complex high- dimensional data. These models are developed in close collaboration with biologists at UC Los Angeles and UC Santa Cruz, incorporating domain-specific biological knowledge from microbiome research, and consequently, they yield biologically interpretable inferences. Our findings, integrated into microbiome research through collaboration, will advance our understanding of how microbes are functionally related to the host, the environment, and other microbes. This understanding can ultimately lead to improvements in human health or the environment through microbiome monitoring or manipulation. Another key aspect of the project involves disseminating the proposed methods through user-friendly software for public use.