Microbial communities are dynamic, evolving through interactions within taxa and in response to environmental changes. Longitudinal studies are increasingly critical for unraveling these complex temporal dynamics, providing insights into microbial functionality and inter-dependencies. However, traditional clustering methods often fail to analyze microbial data due to their inherent sparsity, high dimensionality, and heterogeneity, along with a failure to recognize samples' potential in belonging to multiple clusters. In this context, the probabilistic Latent Dirichlet Allocation (LDA) topic model emerges as a superior alternative. Our study introduces an adaptation of time-aligned LDA designed specifically for longitudinal microbiome analyses. This novel framework focus on aligning microbial topics across sequential time points, thereby addressing the unique challenges of time-variant data. Further, it involves constructing credible intervals from posterior samples and quantitatively delineating the differences in topic proportions across experimental conditions facilitated by a linear mixed model. Applying this framework to gut microbial specimens from pregnant women participating in the Be Healthy in Pregnancy study, we observed enhanced sensitivity in identifying significant temporal dynamics of microbial communities during and after pregnancy compared to the standard LDA.