Using Novel Machine Learning Approaches to Understand Subgingival Microbiome Heterogeneity in Relation to Health and Disease

Project Summary Periodontitis is a complex disease marked by progressive inflammation and destruction of the tissues and bone supporting teeth. Subgingival microbiome dysbiosis is a hallmark of periodontitis with dramatic compositional shifts observed between periodontal health and severe periodontitis. However, there is limited knowledge on the inter-individual variability in subgingival microbiome in large cohorts presenting diverse periodontal phenotypes varying in extent and severity and potentially influenced by systemic comorbidities and exposures. Moreover, most current knowledge on the subgingival microbiome is based on species-level taxonomies, failing to capture strain-level variability which is an important determinant of bacterial virulence. Such detailed understanding of microbiome heterogeneity is needed to elucidate associations of different microbiome states with clinical phenotypes and periodontitis progression. This proposal aims to address these knowledge gaps by leveraging existing population-based cohorts encompassing cross-sectional and 5-year longitudinal follow up datasets and employing advanced machine learning techniques, which enable to capture intricate, nonlinear patterns in microbiome distributions. Our hypothesis is that different microbiome states associated with periodontitis clinical phenotypes and host characteristics exist and can predict periodontitis progression. In Aim 1 (K99 phase), we will use manifold learning techniques to construct microbiome landscape models using cross-sectional 16S data from two large cohorts encompassing over 2,600 subjects. We will identify heterogeneous microbiome states and their relationships with periodontitis phenotypes, and will then examine the predictive ability and dynamics of these microbial states in relation to periodontitis progression over a 5-year period. In Aim 2 (K99/R00 phase), we will refine strain profiling methodologies and establish a strain-level landscape model using metagenomic sequencing to identify variations in strain compositions and functional alterations correlated with periodontitis phenotypes. Upon successful completion of these specific aims, our research is expected to provide a comprehensive understanding of the subgingival microbiome heterogeneity elucidating different microbiome states and their associations with periodontitis offering opportunities for developing targeted, personalized treatment strategies. The application of machine learning techniques is a significant advantage, enabling the analysis of large, complex datasets and the discovery of patterns and relationships that traditional statistical methods can not reveal. The training phase of this proposal will be conducted under the guidance of an experienced mentorship team, equipping Dr. Lu Li with expertise in microbiome analysis, epidemiology and bioinformatics. Additionally, training in scientific communication, grant writing, and leadership skills will be provided, essential for establishing a successful independent research career. The mentorship team will offer a collaborative and intellectually stimulating environment, fostering the development of my scientific career as an independent investigator and ensuring successful project completion.