Multiscale analysis of count data through topic alignment

Sub-communities of the vaginal microbiota in pregnant and non-pregnant women

Diverse and non-Lactobacillus-dominated vaginal microbial communities are associated with adverse health outcomes such as preterm birth and the acquisition of sexually transmitted infections. Despite the importance of recognizing and understanding the key risk-associated features of these communities, their heterogeneous structure and properties remain ill-defined. Clustering approaches are commonly used to characterize vaginal communities, but they lack sensitivity and robustness in resolving substructures and revealing transitions between potential sub-communities. Here, we address this need with an approach based on mixed membership topic models. Using longitudinal data from cohorts of pregnant and non-pregnant study participants, we show that topic models more accurately describe sample composition, longitudinal changes, and better predict the loss of Lactobacillus dominance. We identify several non-Lactobacillus-dominated sub-communities common to both cohorts and independent of reproductive status. In non-pregnant individuals, we find that the menstrual cycle modulates transitions between and within sub-communities, as well as the concentrations of half of the cytokines and 18% of metabolites. Overall, our analyses based on mixed membership models reveal substructures of vaginal ecosystems which may have important clinical and biological associations.

Microbiome subcommunity learning with logistic-tree normal latent Dirichlet allocation

Mixed-membership (MM) models such as latent Dirichlet allocation (LDA) have been applied to microbiome compositional data to identify latent subcommunities of microbial species. These subcommunities are informative for understanding the biological interplay of microbes and for predicting health outcomes. However, microbiome compositions typically display substantial cross-sample heterogeneities in subcommunity compositions-that is, the variability in the proportions of microbes in shared subcommunities across samples-which is not accounted for in prior analyses. As a result, LDA can produce inference, which is highly sensitive to the specification of the number of subcommunities and often divides a single subcommunity into multiple artificial ones. To address this limitation, we incorporate the logistic-tree normal (LTN) model into LDA to form a new MM model. This model allows cross-sample variation in the composition of each subcommunity around some "centroid" composition that defines the subcommunity. Incorporation of auxiliary Pólya-Gamma variables enables a computationally efficient collapsed blocked Gibbs sampler to carry out Bayesian inference under this model. By accounting for such heterogeneity, our new model restores the robustness of the inference in the specification of the number of subcommunities and allows meaningful subcommunities to be identified.