Dalla Lana School of Public Health

The Biostatistics Seminar Series presents …

“Statistical Approaches for Microbiome Data: From Workflows to Bayesian LDA Models” by Dr. Anastasia Teterina, University of Toronto

Abstract: Advances in high-throughput sequencing have generated increasingly complex microbiome datasets, characterized by high dimensionality, sparsity, and compositional constraints. These properties introduce significant challenges for statistical analysis. This talk will first summarize key features of microbiome data and review commonly used analytical workflows, highlighting their strengths, limitations, and considerations for interpretation. The second part will present an alternative approach using latent Dirichlet allocation (LDA), a Bayesian latent variable mixture model that enables dimension reduction and characterization of latent community structures. Applied to cohort microbiome data, LDA provides a flexible framework for modeling temporal dynamics through mixture weights. A real dataset will be used to illustrate the method, and opportunities for further methodological development will be discussed.

Speaker: Dr. A. Teterina is a PhD student in Biostatistics at the University of Toronto, specializing in analytical methodologies for high-dimensional omics data in cohort studies with an emphasis on biologically meaningful interpretability. Her research centers on cohort-based studies, where she explores statistical approaches to address the challenges inherent in sparse, compositional, and longitudinal data structures. Her works spans multiple methodological areas, including longitudinal regression frameworks for differential abundance analysis, Bayesian hierarchical generative models for sparse count data, and analytical approaches for longitudinal compositional datasets.