Novel Statistical Methods for Oral Microbiome Data Analysis

PROJECT SUMMARY The oral cavity harbors a great diversity of microbiota that play an important role in human health. Studies have shown that alterations in the oral microbiome are closely associated with oral diseases such as periodontitis and dental caries, as well as numerous extra-oral disorders such as diabetes and cardiovascular diseases. Large- scale epidemiological studies, including our motivating Oral Infections, Glucose Intolerance and Insulin Resistance Study (ORIGINS), have collected rich data to further investigate the association and mechanistic pathways between the oral microbial communities and health outcomes. However, microbiome data are complex, subject to constraints such as compositionality, zero inflation, overdispersion, taxonomic hierarchy, and high dimensionality. Most existing statistical analyses rely on ad hoc data manipulation and transformation approaches to bypass these constraints, often leading to uninterpretable and irreproducible results. There is a pressing need for more principled statistical methods. This proposal will develop novel association and mediation methods that fully accommodate the unique data features and harness the power of microbiome data. We will build upon a nascent relative-shift regression framework that provides a highly intuitive and interpretable way of modeling compositional data. In Aim 1, we will significantly extend the scope of the relative-shift model by generalizing it to non-Gaussian responses and nolinear relationships. The new methods can better depict the complex association between microbiota and diverse phenotypes. In Aim 2, we will develop novel mediation methods with microbial exposures (Sub-Aim 2a) and microbial mediators (Sub-Aim 2b). The methods will provide a solid foundation for quantifying and testing mediation effects pertaining to microbiome data. The proposed methods will be applied to the ORIGINS data to identify crucial oral microbiota associated with oral/extra-oral phenotypes and to delineate the mediation pathways. The methods and an accompanying software package will also have general utilities in other microbiome studies and promise to enhance our understanding of the microbiome’s critical role in human health.