Collaborative Research Team Projects – Project 29

Statistical Methodologies and Computational Tools to Identify Microbial Correlates of Canadian Bee Gut Health

To sustain the beekeeping economy for Canadian apiculture and agriculture, it is important to research the causes of colony loss and improve honey bee survival. The best approach for understanding this complex problem is statistical analysis of risks. This project will advance the field of compositional data analysis by developing novel and more feasible statistical and computational methods to analyze complex multivariate count (MVC) models.

Research Category:
Region: National
Date: 2025–2028

Why Are New Methods for Analyzing MVC Models Needed?

The Canadian Bee Gut Project (CBGP), led by our collaborator Dr. Emma Allen-Vercoe, was funded by the University of Guelph Food from Thought program (2022-2024). CBGP aims to improve the understanding of the role of microbes found in bee guts on the bees’ health status as well as their overwintering survival.

MVC regression models are often used to correlate the composition of the gut microbiome to a set of covariates. In previous research, we developed the theoretical and methodological foundations for optimizing sparse group lasso (SGL)-MVC regression models under the minorization-maximization (MM) algorithm framework through dominating hyperplane regularization (DHR).

The initial stage of the current project will build upon this work and will use the results to prepare an R package for regularized MVC regression.

The residual diagnostic tools from the initial work will be modified and incorporated in the spatial model setting to develop spatial models that can be generalized to the broader class of MVC mixed models.

In addition, the project will develop tools for cases when observations belong to a set of latent but biologically meaningful subpopulations and to accommodate the spatial/temporal correlations often exhibited by real-world data.

Research Aims and Activities

Our main objective here is to further advance the field of compositional data analysis. We will develop novel and more feasible statistical and computational methods to analyze complex multivariate count models, with the following four specific aims:

1. Development of regularized MVC regression models for associating influential factors with gut microbiome compositions
2. Development of spatial-temporal MVC regression models for bee gut microbiome compositions
3. Development of regularized model-based clustering of gut microbiome compositions
4. Development of a regularized tree-based regression model for associating important influential factors with gut microbiome and associating microbiome compositions to phenotypic outcomes

When taken together, these four aims provide a solid class of regularized MVC models that can be applied to count data exhibiting spatial and/or temporal correlations. The theoretical developments provide a nice interpretation of how to fit these models via a series of iteratively reweighted regularized regressions. We expect our results to generalize to other types of MVC data.